diff --git a/install.ps1 b/install.ps1
index 215c8f6040..f1f16d818d 100644
--- a/install.ps1
+++ b/install.ps1
@@ -681,13 +681,13 @@ shell.Run cmd, 0, False
         if ($SkipTorch) {
             # No-torch: install unsloth + unsloth-zoo with --no-deps, then
             # runtime deps (typer, safetensors, transformers, etc.) with --no-deps.
-            uv pip install --python $VenvPython --no-deps --reinstall-package unsloth --reinstall-package unsloth-zoo "unsloth>=2026.3.14" unsloth-zoo
+            uv pip install --python $VenvPython --no-deps --reinstall-package unsloth --reinstall-package unsloth-zoo "unsloth>=2026.3.16" unsloth-zoo
             $NoTorchReq = Find-NoTorchRuntimeFile
             if ($NoTorchReq) {
                 uv pip install --python $VenvPython --no-deps -r $NoTorchReq
             }
         } else {
-            uv pip install --python $VenvPython --reinstall-package unsloth --reinstall-package unsloth-zoo "unsloth>=2026.3.14" unsloth-zoo
+            uv pip install --python $VenvPython --reinstall-package unsloth --reinstall-package unsloth-zoo "unsloth>=2026.3.16" unsloth-zoo
         }
         if ($StudioLocalInstall) {
             Write-Host "==> Overlaying local repo (editable)..."
@@ -709,7 +709,7 @@ shell.Run cmd, 0, False
         if ($SkipTorch) {
             # No-torch: install unsloth + unsloth-zoo with --no-deps, then
             # runtime deps (typer, safetensors, transformers, etc.) with --no-deps.
-            uv pip install --python $VenvPython --no-deps --upgrade-package unsloth --upgrade-package unsloth-zoo "unsloth>=2026.3.14" unsloth-zoo
+            uv pip install --python $VenvPython --no-deps --upgrade-package unsloth --upgrade-package unsloth-zoo "unsloth>=2026.3.16" unsloth-zoo
             $NoTorchReq = Find-NoTorchRuntimeFile
             if ($NoTorchReq) {
                 uv pip install --python $VenvPython --no-deps -r $NoTorchReq
@@ -719,7 +719,7 @@ shell.Run cmd, 0, False
                 uv pip install --python $VenvPython -e $RepoRoot --no-deps
             }
         } elseif ($StudioLocalInstall) {
-            uv pip install --python $VenvPython --upgrade-package unsloth "unsloth>=2026.3.14" unsloth-zoo
+            uv pip install --python $VenvPython --upgrade-package unsloth "unsloth>=2026.3.16" unsloth-zoo
             Write-Host "==> Overlaying local repo (editable)..."
             uv pip install --python $VenvPython -e $RepoRoot --no-deps
         } else {
@@ -729,7 +729,7 @@ shell.Run cmd, 0, False
         # Fallback: GPU detection failed to produce a URL -- let uv resolve torch
         Write-Host "==> Installing unsloth (this may take a few minutes)..."
         if ($StudioLocalInstall) {
-            uv pip install --python $VenvPython unsloth-zoo "unsloth>=2026.3.14" --torch-backend=auto
+            uv pip install --python $VenvPython unsloth-zoo "unsloth>=2026.3.16" --torch-backend=auto
             Write-Host "==> Overlaying local repo (editable)..."
             uv pip install --python $VenvPython -e $RepoRoot --no-deps
         } else {
diff --git a/install.sh b/install.sh
index 4c960a129e..b50256ea5a 100755
--- a/install.sh
+++ b/install.sh
@@ -968,7 +968,7 @@ if [ "$_MIGRATED" = true ]; then
         # to prevent transitive torch resolution.
         uv pip install --python "$_VENV_PY" --no-deps \
             --reinstall-package unsloth --reinstall-package unsloth-zoo \
-            "unsloth>=2026.3.14" unsloth-zoo
+            "unsloth>=2026.3.16" unsloth-zoo
         _NO_TORCH_RT="$(_find_no_torch_runtime)"
         if [ -n "$_NO_TORCH_RT" ]; then
             uv pip install --python "$_VENV_PY" --no-deps -r "$_NO_TORCH_RT"
@@ -976,7 +976,7 @@ if [ "$_MIGRATED" = true ]; then
     else
         uv pip install --python "$_VENV_PY" \
             --reinstall-package unsloth --reinstall-package unsloth-zoo \
-            "unsloth>=2026.3.14" unsloth-zoo
+            "unsloth>=2026.3.16" unsloth-zoo
     fi
     if [ "$STUDIO_LOCAL_INSTALL" = true ]; then
         echo "==> Overlaying local repo (editable)..."
@@ -998,7 +998,7 @@ elif [ -n "$TORCH_INDEX_URL" ]; then
         # runtime deps (typer, safetensors, transformers, etc.) with --no-deps.
         uv pip install --python "$_VENV_PY" --no-deps \
             --upgrade-package unsloth --upgrade-package unsloth-zoo \
-            "unsloth>=2026.3.14" unsloth-zoo
+            "unsloth>=2026.3.16" unsloth-zoo
         _NO_TORCH_RT="$(_find_no_torch_runtime)"
         if [ -n "$_NO_TORCH_RT" ]; then
             uv pip install --python "$_VENV_PY" --no-deps -r "$_NO_TORCH_RT"
@@ -1009,7 +1009,7 @@ elif [ -n "$TORCH_INDEX_URL" ]; then
         fi
     elif [ "$STUDIO_LOCAL_INSTALL" = true ]; then
         uv pip install --python "$_VENV_PY" \
-            --upgrade-package unsloth "unsloth>=2026.3.14" unsloth-zoo
+            --upgrade-package unsloth "unsloth>=2026.3.16" unsloth-zoo
         echo "==> Overlaying local repo (editable)..."
         uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
     else
@@ -1020,7 +1020,7 @@ else
     # Fallback: GPU detection failed to produce a URL -- let uv resolve torch
     echo "==> Installing unsloth (this may take a few minutes)..."
     if [ "$STUDIO_LOCAL_INSTALL" = true ]; then
-        uv pip install --python "$_VENV_PY" unsloth-zoo "unsloth>=2026.3.14" --torch-backend=auto
+        uv pip install --python "$_VENV_PY" unsloth-zoo "unsloth>=2026.3.16" --torch-backend=auto
         echo "==> Overlaying local repo (editable)..."
         uv pip install --python "$_VENV_PY" -e "$_REPO_ROOT" --no-deps
     else
diff --git a/pyproject.toml b/pyproject.toml
index e2173d6811..97096dc9e5 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -25,6 +25,29 @@ classifiers = [
     "Topic :: Scientific/Engineering :: Artificial Intelligence",
 ]
 dependencies = [
+    "unsloth_zoo>=2026.3.6",
+    "wheel>=0.42.0",
+    "packaging",
+    "torch>=2.4.0,<2.11.0",
+    "torchvision",
+    "numpy",
+    "tqdm",
+    "psutil",
+    "tyro",
+    "protobuf",
+    "xformers>=0.0.27.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
+    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
+    "triton>=3.0.0 ; ('linux' in sys_platform)",
+    "triton-windows ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
+    "sentencepiece>=0.2.0",
+    "datasets>=3.4.1,!=4.0.*,!=4.1.0,<4.4.0",
+    "accelerate>=0.34.1",
+    "peft>=0.18.0,!=0.11.0",
+    "huggingface_hub>=0.34.0",
+    "hf_transfer",
+    "diffusers",
+    "transformers>=4.51.3,!=4.52.0,!=4.52.1,!=4.52.2,!=4.52.3,!=4.53.0,!=4.54.0,!=4.55.0,!=4.55.1,!=4.57.0,!=4.57.4,!=4.57.5,!=5.0.0,!=5.1.0,<=5.3.0",
+    "trl>=0.18.2,!=0.19.0,<=0.24.0",
     "typer",
     "pydantic",
     "pyyaml",
@@ -61,14 +84,11 @@ studio = [
 exclude = ["images*", "tests*", "kernels/moe*"]
 
 [project.optional-dependencies]
-triton = [
-    "triton>=3.0.0 ; ('linux' in sys_platform)",
-    "triton-windows ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-]
-
-huggingfacenotorch = [
+huggingface = [
+    "unsloth_zoo>=2026.3.6",
     "wheel>=0.42.0",
     "packaging",
+    "torchvision",
     "numpy",
     "tqdm",
     "psutil",
@@ -83,257 +103,132 @@ huggingfacenotorch = [
     "diffusers",
     "transformers>=4.51.3,!=4.52.0,!=4.52.1,!=4.52.2,!=4.52.3,!=4.53.0,!=4.54.0,!=4.55.0,!=4.55.1,!=4.57.0,!=4.57.4,!=4.57.5,!=5.0.0,!=5.1.0,<=5.3.0",
     "trl>=0.18.2,!=0.19.0,<=0.24.0",
-    "sentence-transformers",
-]
-huggingface = [
-    "unsloth[huggingfacenotorch]",
-    "unsloth_zoo>=2026.3.6",
-    "torchvision",
-    "unsloth[triton]",
-]
-windows = [
-    "unsloth[huggingface]",
-    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0 ; (sys_platform == 'win32')",
-    "xformers>=0.0.22.post7 ; (sys_platform == 'win32')",
+    "typer",
+    "pydantic",
+    "pyyaml",
+    "nest-asyncio",
 ]
 base = [
     "unsloth[huggingface]",
 ]
 cu118only = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.22.post7%2Bcu118-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.22.post7%2Bcu118-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.22.post7%2Bcu118-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
+    "xformers==0.0.22.post7 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu121only = [
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.22.post7-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.22.post7-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.22.post7-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
+    "xformers==0.0.22.post7 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch211 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.23%2Bcu118-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.23%2Bcu118-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.23%2Bcu118-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
+    "xformers==0.0.23 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu121onlytorch211 = [
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.23-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.23-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.23-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
+    "xformers==0.0.23 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch212 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.23.post1%2Bcu118-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.23.post1%2Bcu118-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.23.post1%2Bcu118-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
+    "xformers==0.0.23.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu121onlytorch212 = [
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.23.post1-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.23.post1-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.23.post1-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
+    "xformers==0.0.23.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch220 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.24%2Bcu118-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.24%2Bcu118-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.24%2Bcu118-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
+    "xformers==0.0.24 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu121onlytorch220 = [
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.24-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.24-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.24-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
+    "xformers==0.0.24 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch230 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.27%2Bcu118-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.27%2Bcu118-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.27%2Bcu118-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.27%2Bcu118-cp312-cp312-manylinux2014_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
+    "xformers==0.0.27 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu121onlytorch230 = [
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.27-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.27-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.27-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.27-cp312-cp312-manylinux2014_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
+    "xformers==0.0.27 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch240 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.27.post2%2Bcu118-cp39-cp39-manylinux2014_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.27.post2%2Bcu118-cp310-cp310-manylinux2014_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.27.post2%2Bcu118-cp311-cp311-manylinux2014_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.27.post2%2Bcu118-cp312-cp312-manylinux2014_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
+    "xformers==0.0.27.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu121onlytorch240 = [
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.28.post1-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.28.post1-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.28.post1-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.28.post1-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
-]
-cu124onlytorch240 = [
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post1-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post1-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post1-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post1-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post1-cp39-cp39-win_amd64.whl ; python_version=='3.9' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post1-cp310-cp310-win_amd64.whl ; python_version=='3.10' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post1-cp311-cp311-win_amd64.whl ; python_version=='3.11' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post1-cp312-cp312-win_amd64.whl ; python_version=='3.12' and (sys_platform == 'win32')",
+    "xformers==0.0.27.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch250 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.28.post2-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.28.post2-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.28.post2-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.28.post2-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
+    "xformers==0.0.28.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu121onlytorch250 = [
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.28.post2-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.28.post2-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.28.post2-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.28.post2-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
+    "xformers==0.0.28.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu124onlytorch250 = [
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post2-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post2-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post2-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post2-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post2-cp39-cp39-win_amd64.whl ; python_version=='3.9' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post2-cp310-cp310-win_amd64.whl ; python_version=='3.10' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post2-cp311-cp311-win_amd64.whl ; python_version=='3.11' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.28.post2-cp312-cp312-win_amd64.whl ; python_version=='3.12' and (sys_platform == 'win32')",
+    "xformers==0.0.28.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch251 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.29.post1-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.29.post1-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.29.post1-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.29.post1-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
+    "xformers==0.0.29.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu121onlytorch251 = [
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.29.post1-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.29.post1-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.29.post1-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu121/xformers-0.0.29.post1-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
+    "xformers==0.0.29.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu124onlytorch251 = [
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post1-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post1-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post1-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post1-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post1-cp39-cp39-win_amd64.whl ; python_version=='3.9' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post1-cp310-cp310-win_amd64.whl ; python_version=='3.10' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post1-cp311-cp311-win_amd64.whl ; python_version=='3.11' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post1-cp312-cp312-win_amd64.whl ; python_version=='3.12' and (sys_platform == 'win32')",
+    "xformers==0.0.29.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch260 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.29.post3-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.29.post3-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.29.post3-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.29.post3-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
+    "xformers==0.0.29.post3 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu124onlytorch260 = [
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post3-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post3-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post3-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post3-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post3-cp39-cp39-win_amd64.whl ; python_version=='3.9' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post3-cp310-cp310-win_amd64.whl ; python_version=='3.10' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post3-cp311-cp311-win_amd64.whl ; python_version=='3.11' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu124/xformers-0.0.29.post3-cp312-cp312-win_amd64.whl ; python_version=='3.12' and (sys_platform == 'win32')",
+    "xformers==0.0.29.post3 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu126onlytorch260 = [
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp39-cp39-win_amd64.whl ; python_version=='3.9' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp310-cp310-win_amd64.whl ; python_version=='3.10' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp311-cp311-win_amd64.whl ; python_version=='3.11' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp312-cp312-win_amd64.whl ; python_version=='3.12' and (sys_platform == 'win32')",
+    "xformers==0.0.29.post3 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch270 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.30-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.30-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.30-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.30-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.30-cp39-cp39-win_amd64.whl ; python_version=='3.9' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.30-cp310-cp310-win_amd64.whl ; python_version=='3.10' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.30-cp311-cp311-win_amd64.whl ; python_version=='3.11' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.30-cp312-cp312-win_amd64.whl ; python_version=='3.12' and (sys_platform == 'win32')",
+    "xformers==0.0.30 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu126onlytorch270 = [
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.30-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.30-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.30-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.30-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.30-cp39-cp39-win_amd64.whl ; python_version=='3.9' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.30-cp310-cp310-win_amd64.whl ; python_version=='3.10' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.30-cp311-cp311-win_amd64.whl ; python_version=='3.11' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.30-cp312-cp312-win_amd64.whl ; python_version=='3.12' and (sys_platform == 'win32')",
+    "xformers==0.0.30 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu128onlytorch270 = [
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.30-cp39-cp39-manylinux_2_28_x86_64.whl ; python_version=='3.9' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.30-cp310-cp310-manylinux_2_28_x86_64.whl ; python_version=='3.10' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.30-cp311-cp311-manylinux_2_28_x86_64.whl ; python_version=='3.11' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.30-cp312-cp312-manylinux_2_28_x86_64.whl ; python_version=='3.12' and ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.30-cp39-cp39-win_amd64.whl ; python_version=='3.9' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.30-cp310-cp310-win_amd64.whl ; python_version=='3.10' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.30-cp311-cp311-win_amd64.whl ; python_version=='3.11' and (sys_platform == 'win32')",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.30-cp312-cp312-win_amd64.whl ; python_version=='3.12' and (sys_platform == 'win32')",
+    "xformers==0.0.30 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch271 = [
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.31.post1-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu118/xformers-0.0.31.post1-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.31.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu126onlytorch271 = [
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.31.post1-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.31.post1-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.31.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu128onlytorch271 = [
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.31.post1-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.31.post1-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.31.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118onlytorch280 = [
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.32.post2-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.32.post2-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.32.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu126onlytorch280 = [
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.32.post2-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.32.post2-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.32.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu128onlytorch280 = [
-    "xformers @ https://download.pytorch.org/whl/cu129/xformers-0.0.32.post2-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu129/xformers-0.0.32.post2-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.32.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu130onlytorch280 = [
 ]
 cu126onlytorch290 = [
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.33.post1-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.33.post1-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.33.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu128onlytorch290 = [
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.33.post1-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.33.post1-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.33.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu130onlytorch290 = [
-    "xformers @ https://download.pytorch.org/whl/cu130/xformers-0.0.33.post1-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu130/xformers-0.0.33.post1-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.33.post1 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu126onlytorch291 = [
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.33.post2-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.33.post2-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.33.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu128onlytorch291 = [
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.33.post2-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.33.post2-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.33.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu130onlytorch291 = [
-    "xformers @ https://download.pytorch.org/whl/cu130/xformers-0.0.33.post2-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu130/xformers-0.0.33.post2-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.33.post2 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu126onlytorch2100 = [
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.34-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.34-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.34 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu128onlytorch2100 = [
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.34-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu128/xformers-0.0.34-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.34 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu130onlytorch2100 = [
-    "xformers @ https://download.pytorch.org/whl/cu130/xformers-0.0.34-cp39-abi3-manylinux_2_28_x86_64.whl ; ('linux' in sys_platform)",
-    "xformers @ https://download.pytorch.org/whl/cu130/xformers-0.0.34-cp39-abi3-win_amd64.whl ; (sys_platform == 'win32')",
+    "xformers==0.0.34 ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
 ]
 cu118 = [
     "unsloth[huggingface]",
@@ -347,22 +242,22 @@ cu121 = [
 ]
 cu118-torch211 = [
     "unsloth[huggingface]",
-    "bitsandbytes==0.45.5",
+    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu118onlytorch211]",
 ]
 cu121-torch211 = [
     "unsloth[huggingface]",
-    "bitsandbytes==0.45.5",
+    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu121onlytorch211]",
 ]
 cu118-torch212 = [
     "unsloth[huggingface]",
-    "bitsandbytes==0.45.5",
+    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu118onlytorch212]",
 ]
 cu121-torch212 = [
     "unsloth[huggingface]",
-    "bitsandbytes==0.45.5",
+    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu121onlytorch212]",
 ]
 cu118-torch220 = [
@@ -432,17 +327,17 @@ cu124-torch251 = [
 ]
 cu118-torch260 = [
     "unsloth[huggingface]",
-    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
+    "bitsandbytes>=0.45.1",
     "unsloth[cu118onlytorch260]",
 ]
 cu124-torch260 = [
     "unsloth[huggingface]",
-    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
+    "bitsandbytes>=0.45.1",
     "unsloth[cu124onlytorch260]",
 ]
 cu126-torch260 = [
     "unsloth[huggingface]",
-    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
+    "bitsandbytes>=0.45.1",
     "unsloth[cu126onlytorch260]",
 ]
 cu118-torch270 = [
@@ -552,16 +447,19 @@ conda = [
 ]
 colab-torch211 = [
     "unsloth[huggingface]",
-    "bitsandbytes==0.45.5",
+    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu121onlytorch211]",
 ]
+flashattention = [
+    "packaging ; ('linux' in sys_platform)",
+    "ninja ; ('linux' in sys_platform)",
+    "flash-attn>=2.6.3 ; ('linux' in sys_platform)",
+]
 colab-ampere-torch211 = [
     "unsloth[huggingface]",
-    "bitsandbytes==0.45.5",
+    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu121onlytorch211]",
-    "packaging",
-    "ninja",
-    "flash-attn>=2.6.3 ; ('linux' in sys_platform)",
+    "unsloth[flashattention]",
 ]
 colab-torch220 = [
     "unsloth[huggingface]",
@@ -572,9 +470,7 @@ colab-ampere-torch220 = [
     "unsloth[huggingface]",
     "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu121onlytorch220]",
-    "packaging",
-    "ninja",
-    "flash-attn>=2.6.3 ; ('linux' in sys_platform)",
+    "unsloth[flashattention]",
 ]
 colab-new = [
     "unsloth_zoo>=2026.3.6",
@@ -592,24 +488,22 @@ colab-new = [
     "hf_transfer",
     "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[triton]",
-    "sentence-transformers",
+    "typer",
+    "pydantic",
+    "pyyaml",
+    "nest-asyncio",
 ]
 colab-no-deps = [
     "accelerate>=0.34.1",
     "trl>=0.18.2,!=0.19.0,<=0.24.0",
     "peft>=0.18.0",
-    "xformers ; ('linux' in sys_platform or sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
+    "xformers<0.0.27",
     "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "protobuf",
 ]
 colab = [
     "unsloth[cu121]",
 ]
-flashattention = [
-    "packaging ; ('linux' in sys_platform)",
-    "ninja ; ('linux' in sys_platform)",
-    "flash-attn>=2.6.3 ; ('linux' in sys_platform)",
-]
 colab-ampere = [
     "unsloth[colab-ampere-torch220]",
     "unsloth[flashattention]",
@@ -628,13 +522,13 @@ cu121-ampere = [
 ]
 cu118-ampere-torch211 = [
     "unsloth[huggingface]",
-    "bitsandbytes==0.45.5",
+    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu118onlytorch211]",
     "unsloth[flashattention]",
 ]
 cu121-ampere-torch211 = [
     "unsloth[huggingface]",
-    "bitsandbytes==0.45.5",
+    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu121onlytorch211]",
     "unsloth[flashattention]",
 ]
@@ -718,19 +612,19 @@ cu124-ampere-torch251 = [
 ]
 cu118-ampere-torch260 = [
     "unsloth[huggingface]",
-    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
+    "bitsandbytes>=0.45.1",
     "unsloth[cu118onlytorch260]",
     "unsloth[flashattention]",
 ]
 cu124-ampere-torch260 = [
     "unsloth[huggingface]",
-    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
+    "bitsandbytes>=0.45.1",
     "unsloth[cu124onlytorch260]",
     "unsloth[flashattention]",
 ]
 cu126-ampere-torch260 = [
     "unsloth[huggingface]",
-    "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
+    "bitsandbytes>=0.45.1",
     "unsloth[cu126onlytorch260]",
     "unsloth[flashattention]",
 ]
@@ -792,7 +686,6 @@ cu130-ampere-torch280 = [
     "unsloth[huggingface]",
     "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu130onlytorch280]",
-    "unsloth[flashattention]",
 ]
 cu126-ampere-torch290 = [
     "unsloth[huggingface]",
@@ -839,347 +732,8 @@ cu130-ampere-torch2100 = [
     "bitsandbytes>=0.45.5,!=0.46.0,!=0.48.0",
     "unsloth[cu130onlytorch2100]",
 ]
-flashattentiontorch260abiFALSEcu12x = [
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiFALSE-cp39-cp39-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.9'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiFALSE-cp310-cp310-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.10'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiFALSE-cp311-cp311-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.11'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiFALSE-cp312-cp312-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.12'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiFALSE-cp313-cp313-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.13'",
-]
-flashattentiontorch260abiTRUEcu12x = [
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiTRUE-cp39-cp39-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.9'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiTRUE-cp310-cp310-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.10'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiTRUE-cp311-cp311-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.11'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiTRUE-cp312-cp312-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.12'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiTRUE-cp313-cp313-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.13'",
-]
-flashattentiontorch250abiFALSEcu12x = [
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiFALSE-cp39-cp39-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.9'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiFALSE-cp310-cp310-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.10'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiFALSE-cp311-cp311-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.11'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiFALSE-cp312-cp312-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.12'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiFALSE-cp313-cp313-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.13'",
-]
-flashattentiontorch250abiTRUEcu12x = [
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiTRUE-cp39-cp39-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.9'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiTRUE-cp310-cp310-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.10'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiTRUE-cp311-cp311-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.11'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiTRUE-cp312-cp312-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.12'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiTRUE-cp313-cp313-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.13'",
-]
-flashattentiontorch240abiFALSEcu12x = [
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiFALSE-cp39-cp39-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.9'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiFALSE-cp310-cp310-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.10'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiFALSE-cp311-cp311-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.11'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiFALSE-cp312-cp312-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.12'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiFALSE-cp313-cp313-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.13'",
-]
-flashattentiontorch240abiTRUEcu12x = [
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiTRUE-cp39-cp39-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.9'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiTRUE-cp310-cp310-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.10'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiTRUE-cp311-cp311-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.11'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiTRUE-cp312-cp312-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.12'",
-    "flash-attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.4cxx11abiTRUE-cp313-cp313-linux_x86_64.whl ; ('linux' in sys_platform) and python_version == '3.13'",
-]
-intelgputorch260 = [
-    "unsloth_zoo[intelgpu]",
-    "unsloth[huggingfacenotorch]",
-
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.2.0-cp39-cp39-linux_x86_64.whl#sha256=147607f190a7d7aa24ba454def5977fbbfec792fdae18e4ed278cfec29b69271 ; ('linux' in sys_platform) and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.2.0-cp310-cp310-linux_x86_64.whl#sha256=23aa423fa1542afc34f67eb3ba8ef20060f6d1b3a4697eaeab22b11c92b30f2b ; ('linux' in sys_platform) and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.2.0-cp311-cp311-linux_x86_64.whl#sha256=bcfa995229bbfd9ffd8d6c8d9f6428d393e876fa6e23ee3c20e3c0d73ca75ca5 ; ('linux' in sys_platform) and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.2.0-cp312-cp312-linux_x86_64.whl#sha256=bd340903d03470708df3442438acb8b7e08087ab9e61fbe349b2872bf9257ab0 ; ('linux' in sys_platform) and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.2.0-cp313-cp313-linux_x86_64.whl#sha256=814dccc8a07159e6eca74bed70091bc8fea2d9dd87b0d91845f9f38cde62f01c ; ('linux' in sys_platform) and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-manylinux_2_24_x86_64.whl ; ('linux' in sys_platform) and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-win_amd64.whl ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.6.0%2Bxpu-cp39-cp39-linux_x86_64.whl#sha256=6a8adf6dc4c089406e8b3a7e58ab57a463bddf9b07130d2576e76eced43e92af ; ('linux' in sys_platform) and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.6.0%2Bxpu-cp310-cp310-linux_x86_64.whl#sha256=ff4561cbf07c83bbccaa0f6e9bb0e6dcf721bacd53c9c43c4eb0e7331b4792f9 ; ('linux' in sys_platform) and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.6.0%2Bxpu-cp311-cp311-linux_x86_64.whl#sha256=12005f66b810ddd3ab93f86c4522bcfdd412cbd27fc9d189b661ff7509bc5e8a ; ('linux' in sys_platform) and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.6.0%2Bxpu-cp312-cp312-linux_x86_64.whl#sha256=c4c5c67625cdacf35765c2b94e61fe166e3c3f4a14521b1212a59ad1b3eb0f2e ; ('linux' in sys_platform) and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.6.0%2Bxpu-cp313-cp313-linux_x86_64.whl#sha256=e6864f7a60a5ecc43d5d38f59a16e5dd132384f73dfd3a697f74944026038f7b ; ('linux' in sys_platform) and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-]
-intel-gpu-torch260 = [
-    "unsloth[intelgputorch260]"
-]
-intelgputorch270 = [
-    "unsloth_zoo[intelgpu]",
-    "unsloth[huggingfacenotorch]",
-
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp39-cp39-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=749a7098492c6a27b356c97149a4a62973b953eae60bc1b6259260974f344913 ; ('linux' in sys_platform) and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=44362e80abd752471a08341093321955b066daa2cfb4810e73b8e3b240850f93 ; ('linux' in sys_platform) and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=faa6b8c945a837a080f641bc8ccc77a98fa66980dcd7e62e715fd853737343fd ; ('linux' in sys_platform) and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=40f6fb65b345dc9a61813abe7ac9a585f2c9808f414d140cc2a5f11f53ee063c ; ('linux' in sys_platform) and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=b22b4c02ec71b4bfc862ae3cdfd2871dc0b05d2b1802f5db2196e0f897d581e9 ; ('linux' in sys_platform) and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp39-cp39-win_amd64.whl#sha256=d4b738d7fa5100c1bd766f91614962828a4810eb57b4df92cd5214a83505a752 ; sys_platform == 'win32' and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp310-cp310-win_amd64.whl#sha256=143fe8a64d807bcdb7d81bbc062816add325570aa160448454ab6ded4a0a17a1 ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp311-cp311-win_amd64.whl#sha256=a8025459ff325d6e3532eb5cf72519db1b178155e7d60aff6c56beb5968fc758 ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp312-cp312-win_amd64.whl#sha256=0dd07e6d5b872e42e48f5ee140e609d4554ca3cc509d5bf509ac232267cf358e ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.3.0-cp313-cp313-win_amd64.whl#sha256=a936a18182d8e065a9933afc9a3ebbffadd38604969f87c493831214539fc027 ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-manylinux_2_24_x86_64.whl ; ('linux' in sys_platform) and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-win_amd64.whl ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp39-cp39-linux_x86_64.whl#sha256=f8ee75e50fcbb37ed5b498299ca2264da99ab278a93fae2358e921e4a6e28273 ; ('linux' in sys_platform) and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp310-cp310-linux_x86_64.whl#sha256=d6fdc342961d98fdcd9d03dfd491a3208bb5f7fbb435841f8f72ce9fdcd2d026 ; ('linux' in sys_platform) and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp311-cp311-linux_x86_64.whl#sha256=74d07f9357df5cf2bf223ad3c84de16346bfaa0504f988fdd5590d3e177e5e86 ; ('linux' in sys_platform) and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp312-cp312-linux_x86_64.whl#sha256=c806d44aa2ca5d225629f6fbc6c994d5deaac2d2cde449195bc8e3522ddd219a ; ('linux' in sys_platform) and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp313-cp313-linux_x86_64.whl#sha256=25d8277b7f01d42e2e014ccbab57a2692b6ec4eff8dcf894eda1b297407cf97a ; ('linux' in sys_platform) and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp39-cp39-win_amd64.whl#sha256=046e85125266ae69c1a0d083e6c092f947ab4b6b41532c16bafe40dbced845df ; sys_platform == 'win32' and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp310-cp310-win_amd64.whl#sha256=9ebaeffb82b0b3e39b6030927d3ebe0eb62a0e9045a3b2d7b0a9e7b15222c0db ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp311-cp311-win_amd64.whl#sha256=356ba66cee127e7e2c942880bd50e03768306a4ea08d358a0f29c6eebfc4bc81 ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp312-cp312-win_amd64.whl#sha256=94739e665d9b4d5cd7af5f517cb6103f6f9fb421c095184609653a24524040f5 ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.7.0%2Bxpu-cp313-cp313-win_amd64.whl#sha256=31df3cb674918e89bc8c532baa331dc84f4430e1f9c0ec379232db44cba78355 ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-]
-intel-gpu-torch270 = [
-    "unsloth[intelgputorch270]"
-]
-intelgputorch280 = [
-    "unsloth_zoo[intelgpu]",
-    "unsloth[huggingfacenotorch]",
-
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp39-cp39-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=ac4d8e33986b1c3c5e48151640539272b2187e83016985853111b46fb82c3c94 ; 'linux' in sys_platform and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=999fef4c1f711092b9d3086525920545df490de476ecebe899ffc777019ae17f ; 'linux' in sys_platform and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=57b09c8c492985ff6a27cd3a22b08e8f7b96b407bd8030967b6efbb9f63b80cf ; 'linux' in sys_platform and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=df4bb3282bac9a3b90231700077110d8680b338416de03c2b7c6133c9b602649 ; 'linux' in sys_platform and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=60da63c99ca827bdcb0df28e0298bf7d066dc607454c6d6176783cb4e79d838b ; 'linux' in sys_platform and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp39-cp39-win_amd64.whl#sha256=64aea8de349f3e2e0ebf4c24b011a8122531fdffda5776edaef45829cc241cf8 ; sys_platform == 'win32' and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp310-cp310-win_amd64.whl#sha256=ae573d255b257fdbed319a3440dc9d0a721e31160ab7f6eba1b2226e6a409a1d ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp311-cp311-win_amd64.whl#sha256=8e0ea4558e5776d8ddab0264310be9b26aee5641bcac0da023537556d4317b86 ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp312-cp312-win_amd64.whl#sha256=4090dde07a4fffc34aaf855701a9db28e9fccb57b368ade520f1a0f8e811c878 ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.4.0-cp313-cp313-win_amd64.whl#sha256=a33d0888f3c8df028a2d028842715837d0049524d6c06b9bb11869890a13601a ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp39-cp39-linux_x86_64.whl ; 'linux' in sys_platform and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp310-cp310-linux_x86_64.whl ; 'linux' in sys_platform and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp311-cp311-linux_x86_64.whl ; 'linux' in sys_platform and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp312-cp312-linux_x86_64.whl ; 'linux' in sys_platform and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp313-cp313-linux_x86_64.whl ; 'linux' in sys_platform and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp39-cp39-win_amd64.whl#sha256=f2f401276892428e4875cf1d8717c5cbab704b16fc594ccf23795e7b16549a99 ; sys_platform == 'win32' and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp310-cp310-win_amd64.whl#sha256=125c60cd59d51b39581a7e9afcd4679bc3a6b8c1f9440b1bb502a23fdd60571e ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp311-cp311-win_amd64.whl#sha256=47f1a57258cd460e80b38b2ed6744e31587ab77a96b4215bf59546cb4bab5cc0 ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp312-cp312-win_amd64.whl#sha256=0937d8943c145a83d9bafc6f80ef28971167817f9eda26066d33f72caf8a6646 ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.8.0%2Bxpu-cp313-cp313-win_amd64.whl#sha256=e034aab1d71760dc80a731531be43673ffe15e99033b82d24e40d2e6d41bd8bf ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-manylinux_2_24_x86_64.whl ; ('linux' in sys_platform) and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-win_amd64.whl ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp39-cp39-manylinux_2_28_x86_64.whl#sha256=6e981c192045fc249c008441179ff237bb00174d818b875b0475730b63f0eaca ; 'linux' in sys_platform and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp310-cp310-manylinux_2_28_x86_64.whl#sha256=e5ba4805969277175ebfd59cc717093528cc6e3ada89ac2725fc7a3c1fee6169 ; 'linux' in sys_platform and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp311-cp311-manylinux_2_28_x86_64.whl#sha256=74c39c144104416bc4c5ad8c26ab0c169dc5cc6be58059e01bc3665dd0ef676f ; 'linux' in sys_platform and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp312-cp312-manylinux_2_28_x86_64.whl#sha256=0acec355b80c3899841184084f365df336c508602812e34a44007b8b60d53af4 ; 'linux' in sys_platform and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp313-cp313-manylinux_2_28_x86_64.whl#sha256=e2109ae773dad27b98ca17681044b4f876563c37f2382b75de3a371399edcff8 ; 'linux' in sys_platform and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp39-cp39-win_amd64.whl#sha256=5f7904e7048d414379bc8c1167260f1e84204f105db2d0a2f9c89e87ce1cf205 ; sys_platform == 'win32' and python_version == '3.9' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp310-cp310-win_amd64.whl#sha256=005fca5e658ca8e37adb63c1a021c84f5e56dfa6cf0d601d89cfe40b9473f79f ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp311-cp311-win_amd64.whl#sha256=c6d030f5361461550c0ff1339b5bca8585fc1e84fda2e64b6184e65a581e4f98 ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp312-cp312-win_amd64.whl#sha256=91aafd61864cdce27461cbec13ddbf28c1bc6494265a1e4b80131c64a3b7d18f ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.23.0%2Bxpu-cp313-cp313-win_amd64.whl#sha256=71dc4a6421742ed1e7f585b04a100ad53615c341fbccfbc255aefb38ea9091da ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-]
-intel-gpu-torch280 = [
-    "unsloth[intelgputorch280]"
-]
-intelgputorch290 = [
-    "unsloth_zoo[intelgpu]",
-    "unsloth[huggingfacenotorch]",
-
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=c169a1de14c19673b17c751290d467fa282fc90fa5da4314b2e5cdab1f553146 ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=013d9dd5d6479bd22983161f462e61c8dbe1d82e6730624a7a8d5945507eaa61 ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=afc8cabfbf7ed51fd278d1e0f88d6afc157b0201bad4b99d681e4d542f9e66d4 ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=0d24c1716088f2764d0d24c64227732195b6a42706c3c5fc89eeb4904bfa0818 ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp310-cp310-win_amd64.whl#sha256=c83ab007311d9cfb6e809ee5a4587d99a9eef4be720b90da4f1aaa68b45139a0 ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp311-cp311-win_amd64.whl#sha256=debf75348da8e8c7166b4d4a9b91d1508bb8d6581e339f79f7604b2e6746bacd ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp312-cp312-win_amd64.whl#sha256=97337a47425f1963a723475bd61037460e84ba01db4f87a1d662c3718ff6c47e ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp313-cp313-win_amd64.whl#sha256=2caf8138695f6abb023ecd02031a2611ba1bf8fff2f19802567cb2fadefe9e87 ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.9.0%2Bxpu-cp310-cp310-linux_x86_64.whl#sha256=5afbe860ce991825a36b75706a523601087e414b77598ef0d9d3d565741c277d ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.9.0%2Bxpu-cp311-cp311-linux_x86_64.whl#sha256=607fe419c32d6e8e0556f745742e7cff1d0babce51f54be890e0c1422359c442 ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.9.0%2Bxpu-cp312-cp312-linux_x86_64.whl#sha256=376bae584d89980b8e59934d248c38d5fa3b7d4687a4df1a19f4bc1d23dcc8c1 ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.9.0%2Bxpu-cp313-cp313-linux_x86_64.whl#sha256=98d6a06dd7fb185874367b18bd609f05f16fdce4142a5980ca94461949965cd2 ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.9.0%2Bxpu-cp310-cp310-win_amd64.whl#sha256=47cc68f631f65bd9c84924d052cd04dec7531023caa85e80345e9c94611c887d ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.9.0%2Bxpu-cp311-cp311-win_amd64.whl#sha256=d56c44ab4818aba57e5c7b628f422d014e0d507427170a771c5be85e308b0bc6 ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.9.0%2Bxpu-cp312-cp312-win_amd64.whl#sha256=18cad93aaff76a01ce73aef6935ece7cfc03344b905592ec731446c44d44592b ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.9.0%2Bxpu-cp313-cp313-win_amd64.whl#sha256=579929cdc10a76800ead41289cac191ea36d1b16f5f501d3fc25607d4375cd83 ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-manylinux_2_24_x86_64.whl ; ('linux' in sys_platform) and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-win_amd64.whl ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.24.0%2Bxpu-cp310-cp310-manylinux_2_28_x86_64.whl#sha256=cbfae2b79b7549fd368c2462fc8e94f8f26cc450782ee72138e908077c09a519 ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.24.0%2Bxpu-cp311-cp311-manylinux_2_28_x86_64.whl#sha256=044fa36ef4b6b43edcd490b75c853fa4b3eb033c2bded29f8fbcf27734713c67 ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.24.0%2Bxpu-cp312-cp312-manylinux_2_28_x86_64.whl#sha256=4b91e4bec1d740a6211f02578a79888550b73f3a4e1383035f8f6d72f587212c ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.24.0%2Bxpu-cp313-cp313-manylinux_2_28_x86_64.whl#sha256=88239e73ca37254bec84f29cd5887e10ff712de7edbbda3fbb3609cd6190d99e ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.24.0%2Bxpu-cp310-cp310-win_amd64.whl#sha256=19c7da8ca767d593e13a88a12bb08d06e34a673f6f26c2f9c191d60e81c02953 ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.24.0%2Bxpu-cp311-cp311-win_amd64.whl#sha256=9bb0d1421c544ac8e2eca5b47daacaf54706dc9139c003aa5e77ee5f355c5931 ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.24.0%2Bxpu-cp312-cp312-win_amd64.whl#sha256=6a5194bc736089606342d48a3f6822829b167617e9495d91d753dd1bd46fda18 ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.24.0%2Bxpu-cp313-cp313-win_amd64.whl#sha256=da47a3ce2bb7f0301a31124668b5908f9b9e92d6241443de15a310ef9632fd83 ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-]
-intel-gpu-torch290 = [
-    "unsloth[intelgputorch290]"
-]
-intelgputorch210 = [
-    "unsloth_zoo[intelgpu]",
-    "unsloth[huggingfacenotorch]",
-
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=c169a1de14c19673b17c751290d467fa282fc90fa5da4314b2e5cdab1f553146 ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=013d9dd5d6479bd22983161f462e61c8dbe1d82e6730624a7a8d5945507eaa61 ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=afc8cabfbf7ed51fd278d1e0f88d6afc157b0201bad4b99d681e4d542f9e66d4 ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl#sha256=0d24c1716088f2764d0d24c64227732195b6a42706c3c5fc89eeb4904bfa0818 ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp310-cp310-win_amd64.whl#sha256=c83ab007311d9cfb6e809ee5a4587d99a9eef4be720b90da4f1aaa68b45139a0 ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp311-cp311-win_amd64.whl#sha256=debf75348da8e8c7166b4d4a9b91d1508bb8d6581e339f79f7604b2e6746bacd ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp312-cp312-win_amd64.whl#sha256=97337a47425f1963a723475bd61037460e84ba01db4f87a1d662c3718ff6c47e ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "pytorch_triton_xpu @ https://download.pytorch.org/whl/pytorch_triton_xpu-3.5.0-cp313-cp313-win_amd64.whl#sha256=2caf8138695f6abb023ecd02031a2611ba1bf8fff2f19802567cb2fadefe9e87 ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.10.0%2Bxpu-cp310-cp310-linux_x86_64.whl#sha256=abb1d1ec1ac672bac0ff35420c965f2df0c636ef9d94e2a830e34578489d0a57 ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.10.0%2Bxpu-cp311-cp311-linux_x86_64.whl#sha256=71ad2f82da0f41eaec159f39fc85854e27c2391efa91b373e550648a6f4aaad3 ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.10.0%2Bxpu-cp312-cp312-linux_x86_64.whl#sha256=b473571d478912f92881cc13f15fa18f8463fb0fb8a068c96ed47a7d45a4da0a ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.10.0%2Bxpu-cp313-cp313-linux_x86_64.whl#sha256=3bc64a746ff25a93de140902c60c9e819d7413f5cea1e88d80999c27a5901e9c ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.10.0%2Bxpu-cp310-cp310-win_amd64.whl#sha256=ce50691ab3fb6301d9b7bb8b3834cf5fa7152a2b5f91fd24c5efdc601a25b780 ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.10.0%2Bxpu-cp311-cp311-win_amd64.whl#sha256=cb9d37f21cb9fb7df67d62863f021c3144e8d8832b9ea8e8523ac308bc620ea1 ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.10.0%2Bxpu-cp312-cp312-win_amd64.whl#sha256=3ad605be4728b6d3a28a44d07dd794b1a9e45551b0057815bf25eb2a6d6a56a7 ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torch @ https://download.pytorch.org/whl/xpu/torch-2.10.0%2Bxpu-cp313-cp313-win_amd64.whl#sha256=2b4b56dd6c792aef82006904fa888692e3782e4ae5da27526801bad4898f05a5 ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-manylinux_2_24_x86_64.whl ; ('linux' in sys_platform) and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "bitsandbytes @ https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-1.33.7.preview-py3-none-win_amd64.whl ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.25.0%2Bxpu-cp310-cp310-manylinux_2_28_x86_64.whl#sha256=7e1e7b170fcf7161c8499b67156c5a05462243626dc0974010791a0bab4378d3 ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.25.0%2Bxpu-cp311-cp311-manylinux_2_28_x86_64.whl#sha256=bd6add201bd7628af70437292e1447abb368e0b5f4ff9abd334ae435efd44792 ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.25.0%2Bxpu-cp312-cp312-manylinux_2_28_x86_64.whl#sha256=6ad2543496bc29e59d3dd614a94d09aa9870318aedb66045344fffddfedd2cf8 ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.25.0%2Bxpu-cp313-cp313-manylinux_2_28_x86_64.whl#sha256=80269f37865fcd8b57f20e4786efae2200bfa2b2727926c3c7acc82f0e7d3548 ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.25.0%2Bxpu-cp310-cp310-win_amd64.whl#sha256=6b9485ba85dcba4d196d6134d9c3332fb228fb2556416bf0450a64e8a472fcba ; sys_platform == 'win32' and python_version == '3.10' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.25.0%2Bxpu-cp311-cp311-win_amd64.whl#sha256=36cbaedf10f6412af5c89afd9aeea474e6a56a0050348ada8fabe1ecaf6b879e ; sys_platform == 'win32' and python_version == '3.11' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.25.0%2Bxpu-cp312-cp312-win_amd64.whl#sha256=738357d97468d75fe3d510ac37e65130f2787f81d9bbc1518898f7396dc3403f ; sys_platform == 'win32' and python_version == '3.12' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-    "torchvision @ https://download.pytorch.org/whl/xpu/torchvision-0.25.0%2Bxpu-cp313-cp313-win_amd64.whl#sha256=1c4b44b36a557f7381e3076fb8843366742238648441d607c8d049c6da0f8886 ; sys_platform == 'win32' and python_version == '3.13' and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-]
-intel-gpu-torch210 = [
-    "unsloth[intelgputorch210]"
-]
-intel = [
-    "unsloth[intelgputorch280]",
-]
-amd = [
-    "unsloth[huggingfacenotorch]",
-    "bitsandbytes>=0.49.1 ; ('linux' in sys_platform) and (platform_machine == 'AMD64' or platform_machine == 'x86_64' or platform_machine == 'aarch64')",
-    "bitsandbytes>=0.49.1 ; (sys_platform == 'win32') and (platform_machine == 'AMD64' or platform_machine == 'x86_64')",
-]
-rocm702-torch280 = [
-    "unsloth[amd]",
-
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/triton-3.4.0%2Brocm7.0.2.gitf9e5bf54-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/triton-3.4.0%2Brocm7.0.2.gitf9e5bf54-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/triton-3.4.0%2Brocm7.0.2.gitf9e5bf54-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torch-2.8.0%2Brocm7.0.2.lw.git245bf6ed-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torch-2.8.0%2Brocm7.0.2.lw.git245bf6ed-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torch-2.8.0%2Brocm7.0.2.lw.git245bf6ed-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torchvision-0.23.0%2Brocm7.0.2.git824e8c87-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torchvision-0.23.0%2Brocm7.0.2.git824e8c87-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.0.2/torchvision-0.23.0%2Brocm7.0.2.git824e8c87-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-]
-rocm72-torch291 = [
-    "unsloth[amd]",
-
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.5.1%2Brocm7.2.0.gita272dfa8-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.5.1%2Brocm7.2.0.gita272dfa8-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.5.1%2Brocm7.2.0.gita272dfa8-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.5.1%2Brocm7.2.0.gita272dfa8-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.9.1%2Brocm7.2.0.lw.git7e1940d4-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.9.1%2Brocm7.2.0.lw.git7e1940d4-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.9.1%2Brocm7.2.0.lw.git7e1940d4-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.9.1%2Brocm7.2.0.lw.git7e1940d4-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/windows/rocm-rel-7.2/torch-2.9.1%2Brocmsdk20260116-cp312-cp312-win_amd64.whl ; sys_platform == 'win32' and python_version == '3.12'",
-
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.24.0%2Brocm7.2.0.gitb919bd0c-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.24.0%2Brocm7.2.0.gitb919bd0c-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.24.0%2Brocm7.2.0.gitb919bd0c-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.24.0%2Brocm7.2.0.gitb919bd0c-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/windows/rocm-rel-7.2/torchvision-0.24.1%2Brocmsdk20260116-cp312-cp312-win_amd64.whl ; sys_platform == 'win32' and python_version == '3.12'",
-]
-rocm711-torch291 = [
-    "unsloth[amd]",
-
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/triton-3.5.1%2Brocm7.1.1.gita272dfa8-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/triton-3.5.1%2Brocm7.1.1.gita272dfa8-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/triton-3.5.1%2Brocm7.1.1.gita272dfa8-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/triton-3.5.1%2Brocm7.1.1.gita272dfa8-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torch-2.9.1%2Brocm7.1.1.lw.git351ff442-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torch-2.9.1%2Brocm7.1.1.lw.git351ff442-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torch-2.9.1%2Brocm7.1.1.lw.git351ff442-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torch-2.9.1%2Brocm7.1.1.lw.git351ff442-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torchvision-0.24.0%2Brocm7.1.1.gitb919bd0c-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torchvision-0.24.0%2Brocm7.1.1.gitb919bd0c-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torchvision-0.24.0%2Brocm7.1.1.gitb919bd0c-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torchvision-0.24.0%2Brocm7.1.1.gitb919bd0c-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-]
-rocm72-torch2100 = [
-    "unsloth[amd]",
-
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.6.0%2Brocm7.2.0.gitba5c1517-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.6.0%2Brocm7.2.0.gitba5c1517-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.6.0%2Brocm7.2.0.gitba5c1517-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/triton-3.6.0%2Brocm7.2.0.gitba5c1517-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.10.0%2Brocm7.2.0.lw.gitb6ee5fde-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.10.0%2Brocm7.2.0.lw.gitb6ee5fde-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.10.0%2Brocm7.2.0.lw.gitb6ee5fde-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torch-2.10.0%2Brocm7.2.0.lw.gitb6ee5fde-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.25.0%2Brocm7.2.0.git82df5f59-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.25.0%2Brocm7.2.0.git82df5f59-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.25.0%2Brocm7.2.0.git82df5f59-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.2/torchvision-0.25.0%2Brocm7.2.0.git82df5f59-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-]
-rocm711-torch2100 = [
-    "unsloth[amd]",
-
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/triton-3.6.0%2Brocm7.1.1.gitba5c1517-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/triton-3.6.0%2Brocm7.1.1.gitba5c1517-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/triton-3.6.0%2Brocm7.1.1.gitba5c1517-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "triton @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/triton-3.6.0%2Brocm7.1.1.gitba5c1517-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torch-2.10.0%2Brocm7.1.1.lw.gitd9556b05-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torch-2.10.0%2Brocm7.1.1.lw.gitd9556b05-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torch-2.10.0%2Brocm7.1.1.lw.gitd9556b05-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torch @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torch-2.10.0%2Brocm7.1.1.lw.gitd9556b05-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torchvision-0.25.0%2Brocm7.1.1.git82df5f59-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.10' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torchvision-0.25.0%2Brocm7.1.1.git82df5f59-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.11' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torchvision-0.25.0%2Brocm7.1.1.git82df5f59-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.12' and platform_machine == 'x86_64'",
-    "torchvision @ https://repo.radeon.com/rocm/manylinux/rocm-rel-7.1.1/torchvision-0.25.0%2Brocm7.1.1.git82df5f59-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and python_version == '3.13' and platform_machine == 'x86_64'",
-]
 
 [project.urls]
 homepage = "https://unsloth.ai"
 documentation = "https://unsloth.ai/docs"
 repository = "https://github.com/unslothai/unsloth"
-
-[tool.ruff]
-target-version = "py311"
-force-exclude = true
-extend-exclude = [
-    "*chat_templates.py",
-    "*ollama_template_mappers.py",
-    "*_auto_install.py",
-    "*mapper.py",
-]
-
-[tool.ruff.lint]
-select = ["E9", "F63", "F7", "F82"]
-ignore = [
-    "E402",
-    "E722",
-    "F403",
-    "F405",
-    "F811",
-    "F821",
-    "F841",
-    "F401",
-    "E731",
-    "E741",
-    "F601",
-    "E712",
-]
-
-[tool.ruff.format]
diff --git a/studio/backend/colab.py b/studio/backend/colab.py
index efd0e10bdb..ab2b656627 100644
--- a/studio/backend/colab.py
+++ b/studio/backend/colab.py
@@ -82,7 +82,6 @@ def start(port: int = 8888):
         from colab import start
         start()
     """
-    import sys
 
     logger.info("🦥 Starting Unsloth Studio...")
 
diff --git a/studio/backend/core/data_recipe/jobs/worker.py b/studio/backend/core/data_recipe/jobs/worker.py
index 63e38bd18d..47af7be3ea 100644
--- a/studio/backend/core/data_recipe/jobs/worker.py
+++ b/studio/backend/core/data_recipe/jobs/worker.py
@@ -4,8 +4,6 @@
 from __future__ import annotations
 
 import json
-import structlog
-import loggers
 import logging
 import re
 import shutil
diff --git a/studio/backend/core/data_recipe/local_callable_validators.py b/studio/backend/core/data_recipe/local_callable_validators.py
index c32b2fccaf..53eee3b584 100644
--- a/studio/backend/core/data_recipe/local_callable_validators.py
+++ b/studio/backend/core/data_recipe/local_callable_validators.py
@@ -5,7 +5,6 @@ from __future__ import annotations
 
 import json
 import os
-import structlog
 import subprocess
 from copy import deepcopy
 from dataclasses import dataclass
diff --git a/studio/backend/core/export/export.py b/studio/backend/core/export/export.py
index 966e045b13..f439556d80 100644
--- a/studio/backend/core/export/export.py
+++ b/studio/backend/core/export/export.py
@@ -8,7 +8,6 @@ Export backend - handles model exporting in various formats
 
 import glob
 import json
-import structlog
 from loggers import get_logger
 import os
 import shutil
@@ -23,7 +22,7 @@ from utils.hardware import clear_gpu_cache
 
 from utils.models import is_vision_model, get_base_model_from_lora
 from utils.models.model_config import detect_audio_type
-from utils.paths import ensure_dir, outputs_root, resolve_export_dir, resolve_output_dir
+from utils.paths import ensure_dir, outputs_root, resolve_export_dir
 from core.inference import get_inference_backend
 
 logger = get_logger(__name__)
diff --git a/studio/backend/core/export/orchestrator.py b/studio/backend/core/export/orchestrator.py
index a9fbe659b3..eb6be0ca97 100644
--- a/studio/backend/core/export/orchestrator.py
+++ b/studio/backend/core/export/orchestrator.py
@@ -15,13 +15,11 @@ Pattern follows core/inference/orchestrator.py.
 """
 
 import atexit
-import structlog
 from loggers import get_logger
 import multiprocessing as mp
 import queue
 import threading
 import time
-from pathlib import Path
 from typing import Any, List, Optional, Tuple
 from utils.paths import outputs_root
 
diff --git a/studio/backend/core/export/worker.py b/studio/backend/core/export/worker.py
index 6af6ff1193..af621b0c66 100644
--- a/studio/backend/core/export/worker.py
+++ b/studio/backend/core/export/worker.py
@@ -17,7 +17,6 @@ Pattern follows core/inference/worker.py and core/training/worker.py.
 
 from __future__ import annotations
 
-import structlog
 from loggers import get_logger
 import os
 import sys
diff --git a/studio/backend/core/inference/audio_codecs.py b/studio/backend/core/inference/audio_codecs.py
index bcf3ec2937..613cdb1d44 100644
--- a/studio/backend/core/inference/audio_codecs.py
+++ b/studio/backend/core/inference/audio_codecs.py
@@ -9,7 +9,6 @@ Supports: SNAC (Orpheus), CSM (Sesame), BiCodec (Spark), DAC (OuteTTS)
 import io
 import re
 import wave
-import structlog
 from loggers import get_logger
 from typing import Optional, Tuple
 
diff --git a/studio/backend/core/inference/inference.py b/studio/backend/core/inference/inference.py
index 1a265690ff..6051090df3 100644
--- a/studio/backend/core/inference/inference.py
+++ b/studio/backend/core/inference/inference.py
@@ -7,21 +7,16 @@ Core inference backend - streamlined
 
 from unsloth import FastLanguageModel, FastVisionModel
 from unsloth.chat_templates import get_chat_template
-from transformers import TextStreamer
 from peft import PeftModel, PeftModelForCausalLM
 
 import json
-import sys
 import torch
 from pathlib import Path
 from typing import Optional, Union, Generator, Tuple
-from utils.models import ModelConfig, get_base_model_from_lora
-from utils.paths import is_model_cached
+from utils.models import ModelConfig
 from utils.utils import format_error_message
 from utils.hardware import get_device, clear_gpu_cache, log_gpu_memory
 from core.inference.audio_codecs import AudioCodecManager
-from io import StringIO
-import structlog
 from loggers import get_logger
 
 
@@ -902,7 +897,6 @@ class InferenceBackend:
         try:
             from utils.datasets import (
                 MODEL_TO_TEMPLATE_MAPPER,
-                get_tokenizer_chat_template,
             )
 
             model_name_lower = self.active_model_name.lower()
@@ -1143,7 +1137,6 @@ class InferenceBackend:
         Uses processor.apply_chat_template with audio embedded in messages (Gemma 3n pattern).
         """
         import threading
-        import numpy as np
 
         model_info = self.models[self.active_model_name]
         model = model_info["model"]
@@ -1736,7 +1729,7 @@ class InferenceBackend:
             formatted_prompt = tokenizer.apply_chat_template(
                 chat_messages, tokenize = False, add_generation_prompt = True
             )
-            logger.info(f"Successfully applied tokenizer's native chat template")
+            logger.info("Successfully applied tokenizer's native chat template")
             return formatted_prompt
         except Exception as e:
             error_msg = str(e).lower()
@@ -1745,7 +1738,7 @@ class InferenceBackend:
                 or "no template argument" in error_msg
             ):
                 logger.info(
-                    f"Base model detected - no built-in chat template available, using fallback formatting"
+                    "Base model detected - no built-in chat template available, using fallback formatting"
                 )
             else:
                 logger.warning(f"Failed to apply tokenizer chat template: {e}")
diff --git a/studio/backend/core/inference/llama_cpp.py b/studio/backend/core/inference/llama_cpp.py
index 05e038dbb7..fb5de094d5 100644
--- a/studio/backend/core/inference/llama_cpp.py
+++ b/studio/backend/core/inference/llama_cpp.py
@@ -12,7 +12,6 @@ import atexit
 import contextlib
 import json
 import struct
-import structlog
 from loggers import get_logger
 import shutil
 import socket
@@ -2034,7 +2033,7 @@ class LlamaCppBackend:
 
         except httpx.ConnectError:
             raise RuntimeError("Lost connection to llama-server")
-        except Exception as e:
+        except Exception:
             if cancel_event is not None and cancel_event.is_set():
                 return
             raise
@@ -2629,7 +2628,7 @@ class LlamaCppBackend:
 
             except httpx.ConnectError:
                 raise RuntimeError("Lost connection to llama-server")
-            except Exception as e:
+            except Exception:
                 if cancel_event is not None and cancel_event.is_set():
                     return
                 raise
@@ -2792,7 +2791,7 @@ class LlamaCppBackend:
 
         except httpx.ConnectError:
             raise RuntimeError("Lost connection to llama-server")
-        except Exception as e:
+        except Exception:
             if cancel_event is not None and cancel_event.is_set():
                 return
             raise
diff --git a/studio/backend/core/inference/orchestrator.py b/studio/backend/core/inference/orchestrator.py
index 6ff7fd2cbf..96f09aefdc 100644
--- a/studio/backend/core/inference/orchestrator.py
+++ b/studio/backend/core/inference/orchestrator.py
@@ -17,7 +17,6 @@ Pattern follows core/training/training.py.
 
 import atexit
 import base64
-import structlog
 from loggers import get_logger
 import multiprocessing as mp
 import queue
@@ -25,7 +24,6 @@ import threading
 import time
 import uuid
 from io import BytesIO
-from pathlib import Path
 from typing import Any, Generator, Optional, Tuple, Union
 
 logger = get_logger(__name__)
diff --git a/studio/backend/core/inference/worker.py b/studio/backend/core/inference/worker.py
index afe0ecc458..edcf809d84 100644
--- a/studio/backend/core/inference/worker.py
+++ b/studio/backend/core/inference/worker.py
@@ -17,7 +17,6 @@ Pattern follows core/training/worker.py.
 from __future__ import annotations
 
 import base64
-import structlog
 from loggers import get_logger
 import os
 import queue as _queue
diff --git a/studio/backend/core/training/trainer.py b/studio/backend/core/training/trainer.py
index 2324916236..490ffbf316 100644
--- a/studio/backend/core/training/trainer.py
+++ b/studio/backend/core/training/trainer.py
@@ -33,22 +33,19 @@ if sys.platform in ("win32", "darwin"):
         sys.path.insert(0, _compile_cache)
 
 import torch
-from utils.hardware import clear_gpu_cache, safe_num_proc, dataset_map_num_proc
+from utils.hardware import clear_gpu_cache, dataset_map_num_proc
 
 torch._dynamo.config.recompile_limit = 64
 from unsloth import FastLanguageModel, FastVisionModel, is_bfloat16_supported
-from unsloth.chat_templates import get_chat_template
 
 import json
 import threading
 import math
-import structlog
 from loggers import get_logger
 import time
 from pathlib import Path
 from typing import Optional, Callable
 from dataclasses import dataclass
-import pandas as pd
 from datasets import Dataset, load_dataset
 
 from utils.models import is_vision_model, detect_audio_type
@@ -814,7 +811,7 @@ class UnslothTrainer:
                 # second attempt because the failed first call's partial
                 # imports clean up the stale state as a side effect.
                 self._source_code_retried = True
-                logger.info(f"\n'could not get source code' — retrying once...\n")
+                logger.info("\n'could not get source code' — retrying once...\n")
                 return self.load_model(
                     model_name = model_name,
                     max_seq_length = max_seq_length,
@@ -1016,7 +1013,7 @@ class UnslothTrainer:
                 # Phase 2: Whisper uses FastModel.get_peft_model with task_type=None
                 from unsloth import FastModel
 
-                logger.info(f"Audio model (whisper) LoRA configuration:")
+                logger.info("Audio model (whisper) LoRA configuration:")
                 logger.info(f"  - Target modules: {target_modules}\n")
 
                 self.model = FastModel.get_peft_model(
@@ -1057,7 +1054,7 @@ class UnslothTrainer:
 
             elif self.is_vlm:
                 # Vision model LoRA
-                logger.info(f"Vision model LoRA configuration:")
+                logger.info("Vision model LoRA configuration:")
                 logger.info(f"  - Finetune vision layers: {finetune_vision_layers}")
                 logger.info(f"  - Finetune language layers: {finetune_language_layers}")
                 logger.info(
@@ -1085,7 +1082,7 @@ class UnslothTrainer:
                 )
             else:
                 # Text model LoRA
-                logger.info(f"Text model LoRA configuration:")
+                logger.info("Text model LoRA configuration:")
                 logger.info(f"  - Target modules: {target_modules}\n")
 
                 self.model = FastLanguageModel.get_peft_model(
@@ -1114,7 +1111,6 @@ class UnslothTrainer:
 
         except Exception as e:
             import traceback
-            import sys
 
             error_details = (
                 f"{type(e).__name__}: {str(e)}"
@@ -1140,7 +1136,6 @@ class UnslothTrainer:
         and strip non-TransformersKwargs params that Unsloth/PEFT inject.
         """
         import types
-        import torch
         import torch.nn as nn
         from transformers.models.csm.modeling_csm import (
             CsmForConditionalGeneration,
@@ -1707,7 +1702,6 @@ class UnslothTrainer:
         """
         import sys
         import torch
-        import numpy as np
         import torchaudio.transforms as T
 
         import subprocess
@@ -2550,7 +2544,7 @@ class UnslothTrainer:
                     custom_format_mapping = custom_format_mapping,
                 )
                 eval_dataset = eval_info["dataset"]
-                logger.info(f"Eval dataset formatted successfully\n")
+                logger.info("Eval dataset formatted successfully\n")
             elif eval_enabled and not has_separate_eval_source:
                 # No separate eval source — split the already-formatted dataset
                 formatted_dataset = dataset_info["dataset"]
@@ -3049,7 +3043,7 @@ class UnslothTrainer:
             else:
                 # Default to warmup_steps if neither provided
                 config_args["warmup_steps"] = 5
-                logger.info(f"Using default warmup_steps: 5\n")
+                logger.info("Using default warmup_steps: 5\n")
 
             # Add save_steps if specified
             save_steps_val = training_args.get("save_steps", 0)
@@ -3191,7 +3185,7 @@ class UnslothTrainer:
                     self.tokenizer, "tokenizer"
                 ):
                     logger.info(
-                        f"  ⚠️ Unwrapping Processor → raw tokenizer for text-only SFTTrainer"
+                        "  ⚠️ Unwrapping Processor → raw tokenizer for text-only SFTTrainer"
                     )
                     sft_tokenizer = self.tokenizer.tokenizer
 
@@ -3499,7 +3493,6 @@ def _ensure_deepseek_ocr_installed():
             sys.path.insert(0, parent_dir)
 
         # Try importing again
-        from deepseek_ocr.modeling_deepseekocr import format_messages
 
         logger.info("DeepSeek OCR module installed successfully")
         logger.info("DeepSeek OCR module installed successfully!\n")
diff --git a/studio/backend/core/training/training.py b/studio/backend/core/training/training.py
index 4439e4e173..22250df850 100644
--- a/studio/backend/core/training/training.py
+++ b/studio/backend/core/training/training.py
@@ -19,12 +19,9 @@ import math
 import multiprocessing as mp
 import queue
 import threading
-import time
-import structlog
 from datetime import datetime, timezone
 from loggers import get_logger
-from dataclasses import dataclass, field
-from pathlib import Path
+from dataclasses import dataclass
 from typing import Optional, Tuple, Any
 
 import matplotlib.pyplot as plt
diff --git a/studio/backend/core/training/worker.py b/studio/backend/core/training/worker.py
index 891dfca8f7..b10b2f0380 100644
--- a/studio/backend/core/training/worker.py
+++ b/studio/backend/core/training/worker.py
@@ -13,7 +13,6 @@ Pattern follows core/data_recipe/jobs/worker.py.
 
 from __future__ import annotations
 
-import structlog
 from loggers import get_logger
 import os
 import platform
@@ -460,7 +459,6 @@ def run_training_process(
             ensure_dir,
             resolve_output_dir,
             resolve_tensorboard_dir,
-            datasets_root,
         )
 
         import transformers
diff --git a/studio/backend/main.py b/studio/backend/main.py
index 65f5e7fe90..09ea1b271a 100644
--- a/studio/backend/main.py
+++ b/studio/backend/main.py
@@ -37,7 +37,7 @@ if os.getenv("ENVIRONMENT_TYPE", "production") == "production":
 from fastapi import Depends, FastAPI, Request
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.staticfiles import StaticFiles
-from fastapi.responses import FileResponse, HTMLResponse, Response
+from fastapi.responses import FileResponse, Response
 from pathlib import Path
 from datetime import datetime
 
@@ -219,7 +219,7 @@ async def get_system_info():
     import platform
     import subprocess
     import psutil
-    from utils.hardware import get_device, get_gpu_memory_info, DeviceType
+    from utils.hardware import get_gpu_memory_info
 
     # GPU Info — query nvidia-smi for physical GPUs, filtered by
     # CUDA_VISIBLE_DEVICES when set (the frontend uses this for GGUF
diff --git a/studio/backend/models/export.py b/studio/backend/models/export.py
index a86596f199..d53ab7d0a0 100644
--- a/studio/backend/models/export.py
+++ b/studio/backend/models/export.py
@@ -6,7 +6,7 @@ Pydantic schemas for Export API.
 """
 
 from pydantic import BaseModel, Field
-from typing import List, Optional, Literal, Dict, Any
+from typing import Optional, Literal, Dict, Any
 
 
 class LoadCheckpointRequest(BaseModel):
diff --git a/studio/backend/routes/auth.py b/studio/backend/routes/auth.py
index db37ed837d..0197f9b717 100644
--- a/studio/backend/routes/auth.py
+++ b/studio/backend/routes/auth.py
@@ -18,7 +18,6 @@ from auth import storage, hashing
 from auth.authentication import (
     create_access_token,
     create_refresh_token,
-    get_current_subject,
     get_current_subject_allow_password_change,
     refresh_access_token,
 )
diff --git a/studio/backend/routes/datasets.py b/studio/backend/routes/datasets.py
index 8333009626..ce99e9d3e0 100644
--- a/studio/backend/routes/datasets.py
+++ b/studio/backend/routes/datasets.py
@@ -12,7 +12,6 @@ import sys
 from pathlib import Path
 from uuid import uuid4
 from fastapi import APIRouter, Depends, HTTPException, UploadFile
-import structlog
 from loggers import get_logger
 
 # Add backend directory to path
diff --git a/studio/backend/routes/export.py b/studio/backend/routes/export.py
index 3e60eaaf20..55f82f6076 100644
--- a/studio/backend/routes/export.py
+++ b/studio/backend/routes/export.py
@@ -7,8 +7,7 @@ Export API routes: checkpoint discovery and model export operations.
 
 import sys
 from pathlib import Path
-from fastapi import APIRouter, Depends, HTTPException, Query
-import structlog
+from fastapi import APIRouter, Depends, HTTPException
 from loggers import get_logger
 
 # Add backend directory to path
diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index 6f44a3c69f..2c17ce21a5 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -13,7 +13,6 @@ from fastapi import APIRouter, Depends, HTTPException, Request
 from fastapi.responses import StreamingResponse, JSONResponse
 from typing import Optional
 import json
-import structlog
 from loggers import get_logger
 import asyncio
 import threading
@@ -82,8 +81,6 @@ from models.inference import (
 )
 from auth.authentication import get_current_subject
 
-import io
-import wave
 import base64
 import numpy as np
 
@@ -328,14 +325,14 @@ async def load_model(
                     training_method = adapter_cfg.get("unsloth_training_method")
                     if training_method == "lora" and load_in_4bit:
                         logger.info(
-                            f"adapter_config.json says unsloth_training_method='lora' — "
-                            f"setting load_in_4bit=False to match 16-bit training"
+                            "adapter_config.json says unsloth_training_method='lora' — "
+                            "setting load_in_4bit=False to match 16-bit training"
                         )
                         load_in_4bit = False
                     elif training_method == "qlora" and not load_in_4bit:
                         logger.info(
-                            f"adapter_config.json says unsloth_training_method='qlora' — "
-                            f"setting load_in_4bit=True to match QLoRA training"
+                            "adapter_config.json says unsloth_training_method='qlora' — "
+                            "setting load_in_4bit=True to match QLoRA training"
                         )
                         load_in_4bit = True
                     elif training_method:
@@ -752,7 +749,6 @@ async def generate_audio(
 
 def _decode_audio_base64(b64: str) -> np.ndarray:
     """Decode base64 audio (any format) → float32 numpy array at 16kHz."""
-    import torch
     import torchaudio
     import tempfile
     import os
diff --git a/studio/backend/routes/models.py b/studio/backend/routes/models.py
index 348ffbf6ea..5b81970042 100644
--- a/studio/backend/routes/models.py
+++ b/studio/backend/routes/models.py
@@ -10,7 +10,6 @@ import sys
 from pathlib import Path
 from fastapi import APIRouter, Body, Depends, HTTPException, Query
 from typing import List, Optional
-import structlog
 from loggers import get_logger
 
 import re as _re
@@ -775,7 +774,6 @@ async def get_gguf_variants(
         # case-insensitive match.
         cached_bytes_by_quant: dict[str, int] = {}
         try:
-            import re as _re
             from huggingface_hub import constants as hf_constants
 
             # Sanitize repo_id: must be "owner/name" with safe chars only
diff --git a/studio/backend/routes/training.py b/studio/backend/routes/training.py
index 4cfb060dee..a072bf3ff4 100644
--- a/studio/backend/routes/training.py
+++ b/studio/backend/routes/training.py
@@ -9,8 +9,7 @@ import sys
 from pathlib import Path
 from fastapi import APIRouter, Depends, HTTPException, Request
 from fastapi.responses import StreamingResponse
-from typing import Dict, Optional, Any
-import structlog
+from typing import Optional, Any
 from loggers import get_logger
 import asyncio
 from datetime import datetime
diff --git a/studio/backend/tests/test_transformers_version.py b/studio/backend/tests/test_transformers_version.py
index f3dae537c7..31f89a24db 100644
--- a/studio/backend/tests/test_transformers_version.py
+++ b/studio/backend/tests/test_transformers_version.py
@@ -4,7 +4,6 @@
 """Tests for transformers version detection with local checkpoint fallbacks."""
 
 import json
-import pytest
 from pathlib import Path
 from unittest.mock import patch
 
diff --git a/studio/backend/tests/test_utils.py b/studio/backend/tests/test_utils.py
index 3c33b33cb3..20ff79d46a 100644
--- a/studio/backend/tests/test_utils.py
+++ b/studio/backend/tests/test_utils.py
@@ -17,7 +17,6 @@ Run with:
     python -m pytest tests/test_utils.py -v
 """
 
-import platform
 from unittest.mock import patch, MagicMock
 
 import pytest
@@ -295,8 +294,6 @@ class TestLogGpuMemory:
             "utilization_pct": 12.5,
             "free_gb": 14.0,
         }
-        import structlog
-        from loggers import get_logger
 
         with (
             patch(
@@ -312,8 +309,6 @@ class TestLogGpuMemory:
 
     def test_logs_cpu_fallback_when_no_gpu(self, caplog):
         fake_info = {"available": False, "backend": "cpu"}
-        import structlog
-        from loggers import get_logger
 
         with (
             patch(
diff --git a/studio/backend/utils/cache_cleanup.py b/studio/backend/utils/cache_cleanup.py
index 4c8e6239a0..50bd599400 100644
--- a/studio/backend/utils/cache_cleanup.py
+++ b/studio/backend/utils/cache_cleanup.py
@@ -12,7 +12,6 @@ by spawned subprocesses.
 """
 
 import shutil
-import structlog
 from loggers import get_logger
 from pathlib import Path
 from typing import List, Optional
diff --git a/studio/backend/utils/datasets/chat_templates.py b/studio/backend/utils/datasets/chat_templates.py
index 35fbaba8f0..f7ea2374f2 100644
--- a/studio/backend/utils/datasets/chat_templates.py
+++ b/studio/backend/utils/datasets/chat_templates.py
@@ -8,14 +8,13 @@ This module contains functions for applying chat templates to datasets
 and generating dataset info summaries.
 """
 
-from .format_detection import detect_dataset_format, detect_multimodal_dataset, detect_custom_format_heuristic
+from .format_detection import detect_custom_format_heuristic
 from .model_mappings import MODEL_TO_TEMPLATE_MAPPER
 from loggers import get_logger
+
 logger = get_logger(__name__)
 
 
-
-
 DEFAULT_ALPACA_TEMPLATE = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
 
 ### Instruction:
@@ -63,18 +62,21 @@ def get_tokenizer_chat_template(tokenizer, model_name):
             )
         except Exception as e:
             logger.info(f"⚠️ Failed to apply Unsloth template '{matched_template}': {e}")
-            logger.info(f"   Falling back to tokenizer's default chat template")
+            logger.info("   Falling back to tokenizer's default chat template")
     else:
         # Check if tokenizer actually has a chat_template set
         has_chat_template = (
-            hasattr(tokenizer, 'chat_template')
-            and tokenizer.chat_template is not None
+            hasattr(tokenizer, "chat_template") and tokenizer.chat_template is not None
         )
         if has_chat_template:
-            logger.info(f"📝 Using tokenizer's own chat template (no Unsloth template match)")
+            logger.info(
+                "📝 Using tokenizer's own chat template (no Unsloth template match)"
+            )
         else:
             # Base model with no chat template — apply default ChatML
-            logger.info(f"📝 No chat template found — applying default ChatML template (base model)")
+            logger.info(
+                "📝 No chat template found — applying default ChatML template (base model)"
+            )
             try:
                 tokenizer = get_chat_template(
                     tokenizer,
@@ -82,7 +84,7 @@ def get_tokenizer_chat_template(tokenizer, model_name):
                 )
             except Exception as e:
                 logger.info(f"⚠️ Failed to apply default ChatML template: {e}")
-                logger.info(f"   Falling back to tokenizer as-is")
+                logger.info("   Falling back to tokenizer as-is")
 
     return tokenizer
 
@@ -99,7 +101,7 @@ def get_dataset_info_summary(dataset_info):
         "sharegpt": "ShareGPT format (needs standardization)",
         "chatml_messages": "ChatML format (messages column) - OpenAI compatible",
         "chatml_conversations": "ChatML format (conversations column) - HuggingFace standard",
-        "unknown": "Unknown format"
+        "unknown": "Unknown format",
     }
 
     return {
@@ -110,7 +112,8 @@ def get_dataset_info_summary(dataset_info):
         "chat_column": dataset_info["chat_column"],
         "is_standardized": dataset_info["is_standardized"],
         "warnings": dataset_info.get("warnings", []),
-        "ready_for_training": dataset_info["is_standardized"] and final_format != "unknown"
+        "ready_for_training": dataset_info["is_standardized"]
+        and final_format != "unknown",
     }
 
 
@@ -154,7 +157,7 @@ def apply_chat_template_to_dataset(
     # Get EOS token if needed
     eos_token = ""
     if add_eos_token:
-        if hasattr(tokenizer, 'eos_token') and tokenizer.eos_token:
+        if hasattr(tokenizer, "eos_token") and tokenizer.eos_token:
             eos_token = tokenizer.eos_token
         else:
             warnings.append("add_eos_token=True but tokenizer has no eos_token")
@@ -167,14 +170,16 @@ def apply_chat_template_to_dataset(
             if not dataset_info.get("auto_detection_attempted", False):
                 custom_format_mapping = detect_custom_format_heuristic(dataset)
                 if custom_format_mapping:
-                    warnings.append(f"Auto-detected column mapping: {custom_format_mapping}")
+                    warnings.append(
+                        f"Auto-detected column mapping: {custom_format_mapping}"
+                    )
                 else:
                     errors.append("Could not auto-detect format mapping")
                     return {
                         "dataset": dataset,
                         "success": False,
                         "warnings": warnings,
-                        "errors": errors
+                        "errors": errors,
                     }
             else:
                 # Already failed once in format_dataset, don't retry
@@ -186,7 +191,7 @@ def apply_chat_template_to_dataset(
                     "dataset": dataset,
                     "success": False,
                     "warnings": warnings,
-                    "errors": errors
+                    "errors": errors,
                 }
 
         if custom_format_mapping:
@@ -209,7 +214,7 @@ def apply_chat_template_to_dataset(
 
                 for i in range(num_examples):
                     convo = []
-                    role_order = ['system', 'user', 'assistant']
+                    role_order = ["system", "user", "assistant"]
 
                     for target_role in role_order:
                         for col_name, role in custom_format_mapping.items():
@@ -218,11 +223,18 @@ def apply_chat_template_to_dataset(
 
                                 if is_user_provided:
                                     # User explicitly mapped - include even if empty
-                                    convo.append({"role": role, "content": str(content) if content else ""})
+                                    convo.append(
+                                        {
+                                            "role": role,
+                                            "content": str(content) if content else "",
+                                        }
+                                    )
                                 else:
                                     # Auto-detected - skip empty
                                     if content and str(content).strip():
-                                        convo.append({"role": role, "content": str(content)})
+                                        convo.append(
+                                            {"role": role, "content": str(content)}
+                                        )
 
                     conversations.append(convo)
 
@@ -232,31 +244,35 @@ def apply_chat_template_to_dataset(
                 return result
 
             try:
-                dataset = dataset.map(_apply_custom_mapping, batched = True, batch_size = batch_size)
+                dataset = dataset.map(
+                    _apply_custom_mapping, batched = True, batch_size = batch_size
+                )
                 # Update to use conversations format
                 final_format = "chatml_conversations"
                 chat_column = "conversations"
                 is_standardized = True
-                warnings.append("Successfully converted to ChatML format via custom mapping")
+                warnings.append(
+                    "Successfully converted to ChatML format via custom mapping"
+                )
             except Exception as e:
                 errors.append(f"Custom format mapping failed: {e}")
                 return {
                     "dataset": dataset,
                     "success": False,
                     "warnings": warnings,
-                    "errors": errors
+                    "errors": errors,
                 }
 
     # ALPACA FORMAT
     if final_format == "alpaca":
-
         # Set alpaca chat template on tokenizer for saving (if not already set)
         # This ensures the template is saved with the model for inference
-        if not (hasattr(tokenizer, 'chat_template') and tokenizer.chat_template):
+        if not (hasattr(tokenizer, "chat_template") and tokenizer.chat_template):
             try:
                 from unsloth.chat_templates import get_chat_template
+
                 tokenizer = get_chat_template(tokenizer, chat_template = "alpaca")
-                logger.info(f"📝 Set alpaca chat template on tokenizer for model saving")
+                logger.info("📝 Set alpaca chat template on tokenizer for model saving")
             except Exception as e:
                 logger.info(f"⚠️ Could not set alpaca template on tokenizer: {e}")
 
@@ -266,12 +282,16 @@ def apply_chat_template_to_dataset(
             for i in range(len(examples["instruction"])):
                 fields = {
                     "instruction": examples["instruction"][i],
-                    "input": examples.get("input", [""] * len(examples["instruction"]))[i],
-                    "output": examples["output"][i]
+                    "input": examples.get("input", [""] * len(examples["instruction"]))[
+                        i
+                    ],
+                    "output": examples["output"][i],
                 }
 
                 try:
-                    text = DEFAULT_ALPACA_TEMPLATE.format(fields["instruction"], fields["input"], fields["output"])
+                    text = DEFAULT_ALPACA_TEMPLATE.format(
+                        fields["instruction"], fields["input"], fields["output"]
+                    )
                     text += eos_token
                     texts.append(text)
                 except KeyError as e:
@@ -284,24 +304,26 @@ def apply_chat_template_to_dataset(
 
         try:
             dataset_map_kwargs = {
-                'batched': True,
-                'batch_size': batch_size,
+                "batched": True,
+                "batch_size": batch_size,
             }
 
             try:
                 from torch.utils.data import IterableDataset
+
                 _is_torch_iterable = isinstance(dataset, IterableDataset)
             except ImportError:
                 _is_torch_iterable = False
 
             if not _is_torch_iterable:
                 from utils.hardware import dataset_map_num_proc
+
                 if num_proc is None or type(num_proc) is not int:
                     num_proc = dataset_map_num_proc()
                 else:
                     num_proc = dataset_map_num_proc(num_proc)
-                dataset_map_kwargs['num_proc'] = num_proc
-                dataset_map_kwargs['desc'] = "Applying template to Alpaca format"
+                dataset_map_kwargs["num_proc"] = num_proc
+                dataset_map_kwargs["desc"] = "Applying template to Alpaca format"
 
             formatted_dataset = dataset.map(formatted_fn, **dataset_map_kwargs)
 
@@ -309,7 +331,7 @@ def apply_chat_template_to_dataset(
                 "dataset": formatted_dataset,
                 "success": True,
                 "warnings": warnings,
-                "errors": errors
+                "errors": errors,
             }
         except Exception as e:
             errors.append(f"Failed to format Alpaca dataset: {e}")
@@ -317,12 +339,11 @@ def apply_chat_template_to_dataset(
                 "dataset": dataset,
                 "success": False,
                 "warnings": warnings,
-                "errors": errors
+                "errors": errors,
             }
 
     # CHATML FORMATS
     elif final_format in ["chatml_messages", "chatml_conversations"]:
-
         if not is_standardized:
             warnings.append("Dataset may not be fully standardized")
 
@@ -337,13 +358,11 @@ def apply_chat_template_to_dataset(
             for convo in convos:
                 try:
                     text = tokenizer.apply_chat_template(
-                        convo,
-                        tokenize = False,
-                        add_generation_prompt = False
+                        convo, tokenize = False, add_generation_prompt = False
                     )
 
                     if remove_bos_prefix:
-                        text = text.removeprefix('<bos>')
+                        text = text.removeprefix("<bos>")
                     text += eos_token
 
                     texts.append(text)
@@ -357,23 +376,25 @@ def apply_chat_template_to_dataset(
         try:
             try:
                 from torch.utils.data import IterableDataset
+
                 _is_torch_iterable = isinstance(dataset, IterableDataset)
             except ImportError:
                 _is_torch_iterable = False
 
             dataset_map_kwargs = {
-                'batched': True,
-                'batch_size': batch_size,
+                "batched": True,
+                "batch_size": batch_size,
             }
 
             if not _is_torch_iterable:
                 from utils.hardware import dataset_map_num_proc
+
                 if num_proc is None or type(num_proc) is not int:
                     num_proc = dataset_map_num_proc()
                 else:
                     num_proc = dataset_map_num_proc(num_proc)
-                dataset_map_kwargs['num_proc'] = num_proc
-                dataset_map_kwargs['desc'] = f"Applying chat template to {final_format}"
+                dataset_map_kwargs["num_proc"] = num_proc
+                dataset_map_kwargs["desc"] = f"Applying chat template to {final_format}"
 
             # Monitor tqdm progress from dataset.map() and relay to callback
             _tqdm_monitor_stop = None
@@ -411,7 +432,7 @@ def apply_chat_template_to_dataset(
                 "dataset": formatted_dataset,
                 "success": True,
                 "warnings": warnings,
-                "errors": errors
+                "errors": errors,
             }
         except Exception as e:
             errors.append(f"Failed to format ChatML dataset: {e}")
@@ -419,7 +440,7 @@ def apply_chat_template_to_dataset(
                 "dataset": dataset,
                 "success": False,
                 "warnings": warnings,
-                "errors": errors
+                "errors": errors,
             }
 
     # UNKNOWN FORMAT
@@ -432,5 +453,5 @@ def apply_chat_template_to_dataset(
             "dataset": dataset,
             "success": False,
             "warnings": warnings,
-            "errors": errors
+            "errors": errors,
         }
diff --git a/studio/backend/utils/datasets/data_collators.py b/studio/backend/utils/datasets/data_collators.py
index 687da74c21..66081e47e3 100644
--- a/studio/backend/utils/datasets/data_collators.py
+++ b/studio/backend/utils/datasets/data_collators.py
@@ -9,7 +9,7 @@ particularly for VLM/OCR processing.
 """
 
 from dataclasses import dataclass
-from typing import Any, List, Optional, Union
+from typing import Any, List
 from loggers import get_logger
 
 logger = get_logger(__name__)
@@ -75,7 +75,6 @@ class DeepSeekOCRDataCollator:
         Returns:
             dict with input_ids, attention_mask, labels, pixel_values, etc.
         """
-        from PIL import Image
 
         # Extract messages and images
         all_messages = []
diff --git a/studio/backend/utils/datasets/dataset_utils.py b/studio/backend/utils/datasets/dataset_utils.py
index fac8c3d295..b5473cdc3a 100644
--- a/studio/backend/utils/datasets/dataset_utils.py
+++ b/studio/backend/utils/datasets/dataset_utils.py
@@ -38,12 +38,7 @@ from .format_conversion import (
 from .chat_templates import (
     apply_chat_template_to_dataset,
     get_dataset_info_summary,
-    get_tokenizer_chat_template,
-    DEFAULT_ALPACA_TEMPLATE,
 )
-from .vlm_processing import generate_smart_vlm_instruction
-from .data_collators import DeepSeekOCRDataCollator, VLMDataCollator
-from .model_mappings import TEMPLATE_TO_MODEL_MAPPER
 from loggers import get_logger
 
 logger = get_logger(__name__)
@@ -701,7 +696,7 @@ def format_dataset(
             }
 
         else:
-            warnings.append(f"Cannot convert unknown format to Alpaca")
+            warnings.append("Cannot convert unknown format to Alpaca")
             return {
                 "dataset": dataset,
                 "detected_format": "unknown",
@@ -766,7 +761,7 @@ def format_dataset(
             }
 
         else:
-            warnings.append(f"Unknown format, attempting standardization")
+            warnings.append("Unknown format, attempting standardization")
             if detected["chat_column"]:
                 try:
                     standardized = standardize_chat_format(
@@ -914,7 +909,7 @@ def format_and_template_dataset(
                         f"falling back to auto-detection"
                     )
                     logger.info(
-                        f"⚠️ User VLM mapping failed, falling back to auto-detection..."
+                        "⚠️ User VLM mapping failed, falling back to auto-detection..."
                     )
                     custom_format_mapping = None  # clear so auto-detection runs below
             else:
diff --git a/studio/backend/utils/datasets/format_conversion.py b/studio/backend/utils/datasets/format_conversion.py
index 39f0113fe0..83f0baef96 100644
--- a/studio/backend/utils/datasets/format_conversion.py
+++ b/studio/backend/utils/datasets/format_conversion.py
@@ -41,7 +41,6 @@ def standardize_chat_format(
     """
     import collections
     import itertools
-    from datasets import IterableDataset
 
     # Check if vision tokenizer is used
     is_vlm = False
diff --git a/studio/backend/utils/datasets/format_detection.py b/studio/backend/utils/datasets/format_detection.py
index 7b70ff3a76..0f686561f6 100644
--- a/studio/backend/utils/datasets/format_detection.py
+++ b/studio/backend/utils/datasets/format_detection.py
@@ -229,7 +229,7 @@ def detect_custom_format_heuristic(dataset):
                     if prefix in ["generation", "pass", "inference"]:
                         return True
 
-        if len(col_lower) <= 2 and not col_lower in ["qa", "q", "a"]:
+        if len(col_lower) <= 2 and col_lower not in ["qa", "q", "a"]:
             return True
 
         return False
diff --git a/studio/backend/utils/datasets/llm_assist.py b/studio/backend/utils/datasets/llm_assist.py
index fdc4f374ab..5b03073987 100644
--- a/studio/backend/utils/datasets/llm_assist.py
+++ b/studio/backend/utils/datasets/llm_assist.py
@@ -19,7 +19,6 @@ import os
 import re
 import textwrap
 import time
-from itertools import islice
 from typing import Any, Optional
 
 from loggers import get_logger
@@ -105,7 +104,7 @@ def precache_helper_gguf():
     finally:
         try:
             enable_progress_bars()
-        except Exception as e:
+        except Exception:
             pass
 
 
diff --git a/studio/backend/utils/hardware/hardware.py b/studio/backend/utils/hardware/hardware.py
index 61ee8a0967..6f5cb607bb 100644
--- a/studio/backend/utils/hardware/hardware.py
+++ b/studio/backend/utils/hardware/hardware.py
@@ -17,7 +17,6 @@ Usage:
 """
 
 import platform
-import structlog
 from loggers import get_logger
 from enum import Enum
 from typing import Optional, Dict, Any
@@ -183,7 +182,6 @@ def get_gpu_memory_info() -> Dict[str, Any]:
     # ---- MLX path (Apple Silicon) ----
     if device == DeviceType.MLX:
         try:
-            import mlx.core as mx
             import psutil
 
             # MLX uses unified memory — report system memory as the pool
diff --git a/studio/backend/utils/inference/inference_config.py b/studio/backend/utils/inference/inference_config.py
index a0377f9869..d2408a3655 100644
--- a/studio/backend/utils/inference/inference_config.py
+++ b/studio/backend/utils/inference/inference_config.py
@@ -13,7 +13,6 @@ from pathlib import Path
 from typing import Dict, Any, Optional
 import json
 import yaml
-import structlog
 from loggers import get_logger
 
 from utils.models.model_config import load_model_defaults
diff --git a/studio/backend/utils/models/checkpoints.py b/studio/backend/utils/models/checkpoints.py
index b6b2e11c2e..81ffcd8eb7 100644
--- a/studio/backend/utils/models/checkpoints.py
+++ b/studio/backend/utils/models/checkpoints.py
@@ -6,7 +6,6 @@ Checkpoint scanning utilities for discovering training runs and their checkpoint
 """
 
 import json
-import structlog
 from loggers import get_logger
 from pathlib import Path
 from typing import List, Optional, Tuple
diff --git a/studio/backend/utils/models/model_config.py b/studio/backend/utils/models/model_config.py
index 5de3fd2cf9..efe7eb1f99 100644
--- a/studio/backend/utils/models/model_config.py
+++ b/studio/backend/utils/models/model_config.py
@@ -18,7 +18,6 @@ from utils.paths import (
     resolve_export_dir,
 )
 from utils.utils import without_hf_auth
-import structlog
 from loggers import get_logger
 import os
 import subprocess
diff --git a/studio/backend/utils/paths/path_utils.py b/studio/backend/utils/paths/path_utils.py
index 1d6a952399..c5758ba306 100644
--- a/studio/backend/utils/paths/path_utils.py
+++ b/studio/backend/utils/paths/path_utils.py
@@ -8,7 +8,6 @@ Path utilities for model and dataset handling
 import os
 from pathlib import Path
 from typing import Optional
-import structlog
 from loggers import get_logger
 
 logger = get_logger(__name__)
diff --git a/studio/backend/utils/transformers_version.py b/studio/backend/utils/transformers_version.py
index d8724de723..6fb0e557c2 100644
--- a/studio/backend/utils/transformers_version.py
+++ b/studio/backend/utils/transformers_version.py
@@ -23,7 +23,6 @@ Strategy:
 
 import importlib
 import json
-import structlog
 from loggers import get_logger
 import os
 import shutil
diff --git a/studio/backend/utils/utils.py b/studio/backend/utils/utils.py
index 4e61a5b969..9bdc7f4517 100644
--- a/studio/backend/utils/utils.py
+++ b/studio/backend/utils/utils.py
@@ -6,7 +6,6 @@ Shared backend utilities
 """
 
 import os
-import structlog
 from loggers import get_logger
 from contextlib import contextmanager
 from pathlib import Path
diff --git a/studio/install_python_stack.py b/studio/install_python_stack.py
index e8ae22f470..a064e9f401 100644
--- a/studio/install_python_stack.py
+++ b/studio/install_python_stack.py
@@ -353,7 +353,7 @@ def pip_install(
             )
             if result.returncode == 0:
                 return
-            print(_red(f"   uv failed, falling back to pip..."))
+            print(_red("   uv failed, falling back to pip..."))
             if result.stdout:
                 print(result.stdout.decode(errors = "replace"))
 
diff --git a/tests/python/test_cross_platform_parity.py b/tests/python/test_cross_platform_parity.py
index 6dd41be9fa..94098b7320 100644
--- a/tests/python/test_cross_platform_parity.py
+++ b/tests/python/test_cross_platform_parity.py
@@ -5,7 +5,6 @@ from __future__ import annotations
 import re
 from pathlib import Path
 
-import pytest
 
 REPO_ROOT = Path(__file__).resolve().parents[2]
 INSTALL_SH = REPO_ROOT / "install.sh"
diff --git a/tests/python/test_e2e_no_torch_sandbox.py b/tests/python/test_e2e_no_torch_sandbox.py
index f36f69201d..de8f84f977 100644
--- a/tests/python/test_e2e_no_torch_sandbox.py
+++ b/tests/python/test_e2e_no_torch_sandbox.py
@@ -21,7 +21,6 @@ from __future__ import annotations
 
 import os
 import shutil
-import signal
 import subprocess
 import sys
 import textwrap
diff --git a/tests/python/test_install_python_stack.py b/tests/python/test_install_python_stack.py
index 16538ae42b..028cfc0ea0 100644
--- a/tests/python/test_install_python_stack.py
+++ b/tests/python/test_install_python_stack.py
@@ -2,13 +2,11 @@
 
 from __future__ import annotations
 
-import importlib
 import os
 import sys
 from pathlib import Path
 from unittest import mock
 
-import pytest
 
 # Add the studio directory so we can import install_python_stack
 STUDIO_DIR = Path(__file__).resolve().parents[2] / "studio"
diff --git a/tests/python/test_no_torch_filtering.py b/tests/python/test_no_torch_filtering.py
index 5c2926a1f1..138914691b 100644
--- a/tests/python/test_no_torch_filtering.py
+++ b/tests/python/test_no_torch_filtering.py
@@ -10,7 +10,6 @@ Covers:
 
 from __future__ import annotations
 
-import importlib
 import os
 import re
 import subprocess
diff --git a/tests/python/test_studio_import_no_torch.py b/tests/python/test_studio_import_no_torch.py
index 5592a282ff..248bacf231 100644
--- a/tests/python/test_studio_import_no_torch.py
+++ b/tests/python/test_studio_import_no_torch.py
@@ -15,7 +15,6 @@ import ast
 import os
 import shutil
 import subprocess
-import sys
 import tempfile
 import textwrap
 from pathlib import Path
diff --git a/tests/saving/language_models/test_merge_4bit_validation.py b/tests/saving/language_models/test_merge_4bit_validation.py
index 343e737710..9669b92a3b 100644
--- a/tests/saving/language_models/test_merge_4bit_validation.py
+++ b/tests/saving/language_models/test_merge_4bit_validation.py
@@ -1,6 +1,6 @@
 from unsloth import FastLanguageModel
 from unsloth.chat_templates import get_chat_template
-from trl import SFTTrainer, SFTConfig
+from trl import SFTTrainer
 from transformers import DataCollatorForSeq2Seq, TrainingArguments
 from datasets import load_dataset
 import torch
diff --git a/tests/saving/language_models/test_save_merged_grpo_model.py b/tests/saving/language_models/test_save_merged_grpo_model.py
index 67b649305a..9625bbaec6 100644
--- a/tests/saving/language_models/test_save_merged_grpo_model.py
+++ b/tests/saving/language_models/test_save_merged_grpo_model.py
@@ -11,7 +11,6 @@ import sys
 from pathlib import Path
 import multiprocessing as mp
 import gc
-from multiprocessing import Queue
 
 REPO_ROOT = Path(__file__).parents[3]
 sys.path.insert(0, str(REPO_ROOT))
diff --git a/tests/saving/non_peft/test_mistral_non_peft.py b/tests/saving/non_peft/test_mistral_non_peft.py
index e03813367d..7f28334f5d 100644
--- a/tests/saving/non_peft/test_mistral_non_peft.py
+++ b/tests/saving/non_peft/test_mistral_non_peft.py
@@ -1,6 +1,4 @@
 from unsloth import FastLanguageModel
-from transformers import AutoModelForCausalLM
-from peft import PeftModel
 from pathlib import Path
 import sys
 import warnings
diff --git a/tests/saving/non_peft/test_whisper_non_peft.py b/tests/saving/non_peft/test_whisper_non_peft.py
index 303d596c85..98af05f176 100644
--- a/tests/saving/non_peft/test_whisper_non_peft.py
+++ b/tests/saving/non_peft/test_whisper_non_peft.py
@@ -1,6 +1,5 @@
-from unsloth import FastLanguageModel, FastModel
-from transformers import AutoModelForCausalLM, WhisperForConditionalGeneration
-from peft import PeftModel
+from unsloth import FastModel
+from transformers import WhisperForConditionalGeneration
 from pathlib import Path
 import sys
 import warnings
diff --git a/tests/saving/vision_models/test_index_file_sharded_model.py b/tests/saving/vision_models/test_index_file_sharded_model.py
index f737169841..420125c441 100644
--- a/tests/saving/vision_models/test_index_file_sharded_model.py
+++ b/tests/saving/vision_models/test_index_file_sharded_model.py
@@ -28,7 +28,7 @@ train_dataset = dataset.select(range(2000))
 # To select the next 200 examples for evaluation
 eval_dataset = dataset.select(range(2000, 2200))
 
-print(f"✅ Dataset loaded successfully!")
+print("✅ Dataset loaded successfully!")
 print(f"   📈 Training samples: {len(train_dataset)}")
 print(f"   📊 Evaluation samples: {len(eval_dataset)}")
 
@@ -110,10 +110,10 @@ try:
         loftq_config = None,  # And LoftQ
     )
     print("✅ LoRA configuration applied successfully!")
-    print(f"   🎯 LoRA rank (r): 16")
-    print(f"   📊 LoRA alpha: 32")
-    print(f"   🔍 Vision layers: Enabled")
-    print(f"   💬 Language layers: Enabled")
+    print("   🎯 LoRA rank (r): 16")
+    print("   📊 LoRA alpha: 32")
+    print("   🔍 Vision layers: Enabled")
+    print("   💬 Language layers: Enabled")
 except Exception as e:
     print(f"❌ Failed to apply LoRA configuration: {e}")
     raise
@@ -165,10 +165,10 @@ try:
         ),
     )
     print("✅ Trainer setup completed!")
-    print(f"   📦 Batch size: 2")
-    print(f"   🔄 Gradient accumulation steps: 4")
-    print(f"   📈 Max training steps: 10")
-    print(f"   🎯 Learning rate: 2e-4")
+    print("   📦 Batch size: 2")
+    print("   🔄 Gradient accumulation steps: 4")
+    print("   📈 Max training steps: 10")
+    print("   🎯 Learning rate: 2e-4")
     print(f"   💾 Precision: {'BF16' if is_bf16_supported() else 'FP16'}")
 except Exception as e:
     print(f"❌ Failed to setup trainer: {e}")
diff --git a/tests/saving/vision_models/test_push_to_hub_merged.py b/tests/saving/vision_models/test_push_to_hub_merged.py
index 74fa058988..1e2b53d8ff 100644
--- a/tests/saving/vision_models/test_push_to_hub_merged.py
+++ b/tests/saving/vision_models/test_push_to_hub_merged.py
@@ -29,7 +29,7 @@ train_dataset = dataset.select(range(2000))
 # To select the next 200 examples for evaluation
 eval_dataset = dataset.select(range(2000, 2200))
 
-print(f"✅ Dataset loaded successfully!")
+print("✅ Dataset loaded successfully!")
 print(f"   📈 Training samples: {len(train_dataset)}")
 print(f"   📊 Evaluation samples: {len(eval_dataset)}")
 
@@ -111,10 +111,10 @@ try:
         loftq_config = None,  # And LoftQ
     )
     print("✅ LoRA configuration applied successfully!")
-    print(f"   🎯 LoRA rank (r): 16")
-    print(f"   📊 LoRA alpha: 32")
-    print(f"   🔍 Vision layers: Enabled")
-    print(f"   💬 Language layers: Enabled")
+    print("   🎯 LoRA rank (r): 16")
+    print("   📊 LoRA alpha: 32")
+    print("   🔍 Vision layers: Enabled")
+    print("   💬 Language layers: Enabled")
 except Exception as e:
     print(f"❌ Failed to apply LoRA configuration: {e}")
     raise
@@ -166,10 +166,10 @@ try:
         ),
     )
     print("✅ Trainer setup completed!")
-    print(f"   📦 Batch size: 2")
-    print(f"   🔄 Gradient accumulation steps: 4")
-    print(f"   📈 Max training steps: 10")
-    print(f"   🎯 Learning rate: 2e-4")
+    print("   📦 Batch size: 2")
+    print("   🔄 Gradient accumulation steps: 4")
+    print("   📈 Max training steps: 10")
+    print("   🎯 Learning rate: 2e-4")
     print(f"   💾 Precision: {'BF16' if is_bf16_supported() else 'FP16'}")
 except Exception as e:
     print(f"❌ Failed to setup trainer: {e}")
diff --git a/tests/saving/vision_models/test_save_merge_qwen2.5vl32B_model_ocr_benchmark.py b/tests/saving/vision_models/test_save_merge_qwen2.5vl32B_model_ocr_benchmark.py
index ebe078c73b..33f76c55c3 100644
--- a/tests/saving/vision_models/test_save_merge_qwen2.5vl32B_model_ocr_benchmark.py
+++ b/tests/saving/vision_models/test_save_merge_qwen2.5vl32B_model_ocr_benchmark.py
@@ -2,9 +2,6 @@
 
 from unsloth import FastVisionModel
 
-import torch
-from qwen_vl_utils import process_vision_info
-import os
 from datasets import load_dataset
 from trl import SFTTrainer, SFTConfig
 
@@ -20,7 +17,6 @@ from tests.utils.ocr_eval import OCRModelEvaluator
 
 
 ## Dataset Preparation
-from datasets import load_dataset
 
 dataset = load_dataset("lbourdois/OCR-liboaccn-OPUS-MIT-5M-clean", "en", split = "train")
 # To select the first 2000 examples
@@ -66,12 +62,6 @@ train_dataset = [format_data(sample) for sample in train_dataset]
 eval_dataset = [format_data(sample) for sample in eval_dataset]
 
 ## Setup OCR main evaluation function and helpers
-import os
-import torch
-from tqdm import tqdm
-import pandas as pd
-from jiwer import wer, cer
-from qwen_vl_utils import process_vision_info
 
 #
 ocr_evaluator = OCRModelEvaluator()
diff --git a/tests/saving/vision_models/test_save_merge_vision_model_ocr_benchmark.py b/tests/saving/vision_models/test_save_merge_vision_model_ocr_benchmark.py
index b99785bcb1..5722b7697a 100644
--- a/tests/saving/vision_models/test_save_merge_vision_model_ocr_benchmark.py
+++ b/tests/saving/vision_models/test_save_merge_vision_model_ocr_benchmark.py
@@ -2,9 +2,6 @@
 
 from unsloth import FastVisionModel
 
-import torch
-from qwen_vl_utils import process_vision_info
-import os
 from datasets import load_dataset
 from trl import SFTTrainer, SFTConfig
 
@@ -20,7 +17,6 @@ from tests.utils.ocr_eval import OCRModelEvaluator
 
 
 ## Dataset Preparation
-from datasets import load_dataset
 
 dataset = load_dataset("lbourdois/OCR-liboaccn-OPUS-MIT-5M-clean", "en", split = "train")
 # To select the first 2000 examples
@@ -66,12 +62,6 @@ train_dataset = [format_data(sample) for sample in train_dataset]
 eval_dataset = [format_data(sample) for sample in eval_dataset]
 
 ## Setup OCR main evaluation function and helpers
-import os
-import torch
-from tqdm import tqdm
-import pandas as pd
-from jiwer import wer, cer
-from qwen_vl_utils import process_vision_info
 
 #
 ocr_evaluator = OCRModelEvaluator()
diff --git a/tests/studio/install/test_install_llama_prebuilt_logic.py b/tests/studio/install/test_install_llama_prebuilt_logic.py
index eb30ac2745..56d30095f1 100644
--- a/tests/studio/install/test_install_llama_prebuilt_logic.py
+++ b/tests/studio/install/test_install_llama_prebuilt_logic.py
@@ -1,6 +1,5 @@
 import importlib.util
 import io
-import json
 import os
 import sys
 import tarfile
diff --git a/tests/studio/install/test_pr4562_bugfixes.py b/tests/studio/install/test_pr4562_bugfixes.py
index 9b8c6219de..61b1c1571e 100644
--- a/tests/studio/install/test_pr4562_bugfixes.py
+++ b/tests/studio/install/test_pr4562_bugfixes.py
@@ -14,13 +14,11 @@ Run: pytest tests/studio/install/test_pr4562_bugfixes.py -v
 """
 
 import importlib.util
-import json
 import os
 import subprocess
 import sys
 import textwrap
 from pathlib import Path
-from unittest.mock import patch
 
 import pytest
 
diff --git a/tests/test_raw_text.py b/tests/test_raw_text.py
index 9f2e8cda4e..e580e97b69 100644
--- a/tests/test_raw_text.py
+++ b/tests/test_raw_text.py
@@ -7,7 +7,6 @@ Tests basic functionality without heavy dependencies.
 import sys
 import os
 import tempfile
-from pathlib import Path
 import importlib.util
 
 
diff --git a/tests/utils/aime_eval.py b/tests/utils/aime_eval.py
index 131da3e50b..6a80b16ce8 100644
--- a/tests/utils/aime_eval.py
+++ b/tests/utils/aime_eval.py
@@ -207,7 +207,7 @@ def evaluate_model_aime(
 
     print(f"\n{'='*70}")
     print(f"🧮 AIME EVALUATION - {model_type.upper()} MODEL")
-    print(f"Combined Dataset: test2024 + test2025-I + test2025-II")
+    print("Combined Dataset: test2024 + test2025-I + test2025-II")
     print(f"{'='*70}")
 
     # Load combined AIME dataset
@@ -244,7 +244,7 @@ def evaluate_model_aime(
         seed = seed,
     )
 
-    print(f"\n🔧 Configuration:")
+    print("\n🔧 Configuration:")
     print(f"   Temperature: {temperature}")
     print(f"   Samples per question: {n_sampling}")
     print(f"   Max tokens: {max_tokens}")
@@ -421,28 +421,28 @@ def evaluate_model_aime(
     print(f"📊 AIME EVALUATION RESULTS - {model_type.upper()}")
     print(f"{'='*70}")
 
-    print(f"\n🎯 Overall Performance:")
+    print("\n🎯 Overall Performance:")
     print(f"   Total problems:       {total_problems:>6}")
     print(
         f"   Correct answers:      {correct_answers:>6}/{total_problems} ({accuracy:>5.1f}%)"
     )
     print(f"   Pass@{n_sampling}:              {pass_at_k:>10.1f}%")
 
-    print(f"\n📈 Performance by Dataset:")
+    print("\n📈 Performance by Dataset:")
     for source, stats in source_stats.items():
         source_acc = source_accuracies[source]
         print(
             f"   {source:>12}: {stats['correct']:>3}/{stats['total']:>3} ({source_acc:>5.1f}%)"
         )
 
-    print(f"\n🔧 Configuration:")
+    print("\n🔧 Configuration:")
     print(f"   Temperature:          {temperature}")
     print(f"   Samples per problem:  {n_sampling}")
     print(f"   Max tokens:           {max_tokens}")
     print(f"   Top-p:                {top_p}")
     print(f"   Seed:                 {seed}")
 
-    print(f"\n📝 Token Statistics:")
+    print("\n📝 Token Statistics:")
     print(f"   Avg input tokens:     {results['avg_input_tokens']:>10.1f}")
     print(f"   Avg output tokens:    {results['avg_output_tokens']:>10.1f}")
     print(f"   Max input tokens:     {results['max_input_tokens']:>10}")
diff --git a/tests/utils/hf_utils.py b/tests/utils/hf_utils.py
index 8ad6d5ad08..02b0192748 100644
--- a/tests/utils/hf_utils.py
+++ b/tests/utils/hf_utils.py
@@ -12,9 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import os
-from contextlib import contextmanager, nullcontext
-from typing import Callable, Optional
+from contextlib import nullcontext
+from typing import Callable
 
 import bitsandbytes as bnb
 import torch
diff --git a/tests/utils/ocr_eval.py b/tests/utils/ocr_eval.py
index 3c5cd74a22..eff316a094 100644
--- a/tests/utils/ocr_eval.py
+++ b/tests/utils/ocr_eval.py
@@ -348,7 +348,7 @@ class OCRModelEvaluator:
         plt.savefig("ocr_model_comparison.png")
         plt.show()
 
-        print(f"\nVisualization saved to ocr_model_comparison.png")
+        print("\nVisualization saved to ocr_model_comparison.png")
 
     def get_comparison_results(self) -> Dict[str, Dict[str, float]]:
         """Get the current comparison results."""
diff --git a/tests/utils/os_utils.py b/tests/utils/os_utils.py
index 448f13b8a0..a76627946f 100644
--- a/tests/utils/os_utils.py
+++ b/tests/utils/os_utils.py
@@ -98,10 +98,10 @@ def require_package(package_name, executable_name = None):
         for pm_name, cmd in install_commands.items():
             print(f"  {pm_name}: {cmd}")
 
-    print(f"\nAlternatively, install with conda:")
+    print("\nAlternatively, install with conda:")
     print(f"  conda install -c conda-forge {package_name}")
 
-    print(f"\nPlease install the required package and run the script again.")
+    print("\nPlease install the required package and run the script again.")
     sys.exit(1)
 
 
@@ -120,9 +120,9 @@ def require_python_package(package_name, import_name = None, pip_name = None):
         print(f"❌ Error: Python package '{package_name}' is not installed")
         print(f"\nPlease install {package_name} using pip:")
         print(f"  pip install {pip_name}")
-        print(f"  # or with conda:")
+        print("  # or with conda:")
         print(f"  conda install {pip_name}")
-        print(f"\nAfter installation, run this script again.")
+        print("\nAfter installation, run this script again.")
         sys.exit(1)
     else:
         print(f"✓ Python package '{package_name}' is installed")
diff --git a/tests/utils/test_q_galore.py b/tests/utils/test_q_galore.py
index 6dea5014a0..afd4e8c869 100644
--- a/tests/utils/test_q_galore.py
+++ b/tests/utils/test_q_galore.py
@@ -14,7 +14,6 @@
 #
 # Tests for Q-GaLore integration (unsloth/optimizers/).
 
-import pytest
 import sys
 import os
 import torch
diff --git a/unsloth-cli.py b/unsloth-cli.py
index 612da11eb2..3675d2ed02 100644
--- a/unsloth-cli.py
+++ b/unsloth-cli.py
@@ -101,8 +101,6 @@ def run(args):
         return {"text": texts}
 
     def load_dataset_smart(args):
-        from transformers.utils import strtobool
-
         if args.raw_text_file:
             # Use raw text loader
             loader = RawTextDataLoader(tokenizer, args.chunk_size, args.stride)
diff --git a/unsloth/__init__.py b/unsloth/__init__.py
index f922a9fa44..9ec20ca8d4 100644
--- a/unsloth/__init__.py
+++ b/unsloth/__init__.py
@@ -12,9 +12,15 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import warnings, importlib, sys
+import warnings
+import importlib
+import sys
 from packaging.version import Version
-import os, re, subprocess, inspect, functools
+import os
+import re
+import subprocess
+import inspect
+import functools
 import numpy as np
 
 # Log Unsloth is being used
@@ -105,7 +111,7 @@ try:
     import unsloth_zoo
 except PackageNotFoundError:
     raise ImportError(
-        f"Unsloth: Please install unsloth_zoo via `pip install unsloth_zoo` then retry!"
+        "Unsloth: Please install unsloth_zoo via `pip install unsloth_zoo` then retry!"
     )
 except:
     raise
diff --git a/unsloth/_auto_install.py b/unsloth/_auto_install.py
index f6deefeb33..73440595cd 100644
--- a/unsloth/_auto_install.py
+++ b/unsloth/_auto_install.py
@@ -12,32 +12,60 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-try: import torch
-except: raise ImportError('Install torch via `pip install torch`')
+try:
+    import torch
+except:
+    raise ImportError("Install torch via `pip install torch`")
 from packaging.version import Version as V
 import re
+
 v = V(re.match(r"[0-9\.]{3,}", torch.__version__).group(0))
 cuda = str(torch.version.cuda)
 is_ampere = torch.cuda.get_device_capability()[0] >= 8
 USE_ABI = torch._C._GLIBCXX_USE_CXX11_ABI
-if cuda not in ("11.8", "12.1", "12.4", "12.6", "12.8", "13.0"): raise RuntimeError(f"CUDA = {cuda} not supported!")
-if   v <= V('2.1.0'): raise RuntimeError(f"Torch = {v} too old!")
-elif v <= V('2.1.1'): x = 'cu{}{}-torch211'
-elif v <= V('2.1.2'): x = 'cu{}{}-torch212'
-elif v  < V('2.3.0'): x = 'cu{}{}-torch220'
-elif v  < V('2.4.0'): x = 'cu{}{}-torch230'
-elif v  < V('2.5.0'): x = 'cu{}{}-torch240'
-elif v  < V('2.5.1'): x = 'cu{}{}-torch250'
-elif v <= V('2.5.1'): x = 'cu{}{}-torch251'
-elif v  < V('2.7.0'): x = 'cu{}{}-torch260'
-elif v  < V('2.7.9'): x = 'cu{}{}-torch270'
-elif v  < V('2.8.0'): x = 'cu{}{}-torch271'
-elif v  < V('2.8.9'): x = 'cu{}{}-torch280'
-elif v  < V('2.9.1'): x = 'cu{}{}-torch290'
-elif v  < V('2.9.2'): x = 'cu{}{}-torch291'
-elif v  < V('2.10.1'): x = 'cu{}{}-torch2100'
-else: raise RuntimeError(f"Torch = {v} too new!")
-if v > V('2.6.9') and cuda not in ("11.8", "12.6", "12.8", "13.0"): raise RuntimeError(f"CUDA = {cuda} not supported!")
-if v >= V('2.10.0') and cuda not in ("12.6", "12.8", "13.0"): raise RuntimeError(f"Torch 2.10 requires CUDA 12.6, 12.8, or 13.0! Got CUDA = {cuda}")
-x = x.format(cuda.replace(".", ""), "-ampere" if False else "") # is_ampere is broken due to flash-attn
-print(f'pip install --upgrade pip && pip install --no-deps git+https://github.com/unslothai/unsloth-zoo.git && pip install "unsloth[{x}] @ git+https://github.com/unslothai/unsloth.git" --no-build-isolation')
\ No newline at end of file
+if cuda not in ("11.8", "12.1", "12.4", "12.6", "12.8", "13.0"):
+    raise RuntimeError(f"CUDA = {cuda} not supported!")
+if v <= V("2.1.0"):
+    raise RuntimeError(f"Torch = {v} too old!")
+elif v <= V("2.1.1"):
+    x = "cu{}{}-torch211"
+elif v <= V("2.1.2"):
+    x = "cu{}{}-torch212"
+elif v < V("2.3.0"):
+    x = "cu{}{}-torch220"
+elif v < V("2.4.0"):
+    x = "cu{}{}-torch230"
+elif v < V("2.5.0"):
+    x = "cu{}{}-torch240"
+elif v < V("2.5.1"):
+    x = "cu{}{}-torch250"
+elif v <= V("2.5.1"):
+    x = "cu{}{}-torch251"
+elif v < V("2.7.0"):
+    x = "cu{}{}-torch260"
+elif v < V("2.7.9"):
+    x = "cu{}{}-torch270"
+elif v < V("2.8.0"):
+    x = "cu{}{}-torch271"
+elif v < V("2.8.9"):
+    x = "cu{}{}-torch280"
+elif v < V("2.9.1"):
+    x = "cu{}{}-torch290"
+elif v < V("2.9.2"):
+    x = "cu{}{}-torch291"
+elif v < V("2.10.1"):
+    x = "cu{}{}-torch2100"
+else:
+    raise RuntimeError(f"Torch = {v} too new!")
+if v > V("2.6.9") and cuda not in ("11.8", "12.6", "12.8", "13.0"):
+    raise RuntimeError(f"CUDA = {cuda} not supported!")
+if v >= V("2.10.0") and cuda not in ("12.6", "12.8", "13.0"):
+    raise RuntimeError(
+        f"Torch 2.10 requires CUDA 12.6, 12.8, or 13.0! Got CUDA = {cuda}"
+    )
+x = x.format(
+    cuda.replace(".", ""), "-ampere" if False else ""
+)  # is_ampere is broken due to flash-attn
+print(
+    f'pip install --upgrade pip && pip install --no-deps git+https://github.com/unslothai/unsloth-zoo.git && pip install "unsloth[{x}] @ git+https://github.com/unslothai/unsloth.git" --no-build-isolation'
+)
diff --git a/unsloth/chat_templates.py b/unsloth/chat_templates.py
index 35eb871529..0ea25cbcc8 100644
--- a/unsloth/chat_templates.py
+++ b/unsloth/chat_templates.py
@@ -17,13 +17,11 @@ __all__ = [
     "test_chat_templates",
     "test_hf_gguf_equivalence",
     "remove_special_tokens",
-
     "to_sharegpt",
     "standardize_sharegpt",
     "standardize_data_formats",
     "apply_chat_template",
     "train_on_responses_only",
-
     "test_construct_chat_template",
 ]
 
@@ -32,7 +30,6 @@ from torch import LongTensor, FloatTensor
 from transformers.models.llama.modeling_llama import logger
 from .save import patch_saving_functions
 import os
-import shutil
 from .tokenizer_utils import *
 from .models._utils import patch_tokenizer
 import re
@@ -41,272 +38,336 @@ from unsloth_zoo.dataset_utils import (
     train_on_responses_only,
     standardize_data_formats,
 )
+
 standardize_sharegpt = standardize_data_formats
 CHAT_TEMPLATES = {}
 DEFAULT_SYSTEM_MESSAGE = {}
+
+
 def _ollama_template(name: str):
     return OLLAMA_TEMPLATES[name]
 
+
 # =========================================== Unsloth
 # Unsloth efficient template leverages from Zephyr
-unsloth_template = \
-    "{{ bos_token }}"\
-    "{% if messages[0]['role'] == 'system' %}"\
-        "{{ messages[0]['content'] + '\n' }}"\
-        "{% set loop_messages = messages[1:] %}"\
-    "{% else %}"\
-        "{{ '{system_message}' + '\n' }}"\
-        "{% set loop_messages = messages %}"\
-    "{% endif %}"\
-    "{% for message in loop_messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{ '>>> User: ' + message['content'] + '\n' }}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{ '>>> Assistant: ' + message['content'] + eos_token + '\n' }}"\
-        "{% else %}"\
-            "{{ raise_exception('Only user and assistant roles are supported!') }}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ '>>> Assistant: ' }}"\
+unsloth_template = (
+    "{{ bos_token }}"
+    "{% if messages[0]['role'] == 'system' %}"
+    "{{ messages[0]['content'] + '\n' }}"
+    "{% set loop_messages = messages[1:] %}"
+    "{% else %}"
+    "{{ '{system_message}' + '\n' }}"
+    "{% set loop_messages = messages %}"
     "{% endif %}"
+    "{% for message in loop_messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{ '>>> User: ' + message['content'] + '\n' }}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{ '>>> Assistant: ' + message['content'] + eos_token + '\n' }}"
+    "{% else %}"
+    "{{ raise_exception('Only user and assistant roles are supported!') }}"
+    "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ '>>> Assistant: ' }}"
+    "{% endif %}"
+)
 
 unsloth_ollama = _ollama_template("unsloth")
 
 unsloth_eos_token = "eos_token"
-CHAT_TEMPLATES["unsloth"] = (unsloth_template, unsloth_eos_token, False, unsloth_ollama,)
+CHAT_TEMPLATES["unsloth"] = (
+    unsloth_template,
+    unsloth_eos_token,
+    False,
+    unsloth_ollama,
+)
 DEFAULT_SYSTEM_MESSAGE["unsloth"] = "You are a helpful assistant to the user"
 
 # =========================================== Zephyr
 # Zephyr has no BOS!
-zephyr_template = \
-    "{% for message in messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{ '<|user|>\n' + message['content'] + eos_token + '\n' }}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{ '<|assistant|>\n' + message['content'] + eos_token + '\n' }}"\
-        "{% else %}"\
-            "{{ '<|system|>\n' + message['content'] + eos_token + '\n' }}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ '<|assistant|>\n' }}"\
+zephyr_template = (
+    "{% for message in messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{ '<|user|>\n' + message['content'] + eos_token + '\n' }}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{ '<|assistant|>\n' + message['content'] + eos_token + '\n' }}"
+    "{% else %}"
+    "{{ '<|system|>\n' + message['content'] + eos_token + '\n' }}"
     "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ '<|assistant|>\n' }}"
+    "{% endif %}"
+)
 
 zephyr_ollama = _ollama_template("zephyr")
 
 zephyr_eos_token = "eos_token"
-CHAT_TEMPLATES["zephyr"] = (zephyr_template, zephyr_eos_token, False, zephyr_ollama,)
-DEFAULT_SYSTEM_MESSAGE["zephyr"] = None # No system message in Zephyr
+CHAT_TEMPLATES["zephyr"] = (
+    zephyr_template,
+    zephyr_eos_token,
+    False,
+    zephyr_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["zephyr"] = None  # No system message in Zephyr
 
 # =========================================== ChatML
 # ChatML has no BOS and not EOS! Rather <|im_start|> and <|im_end|> acts as BOS / EOS.
-chatml_template = \
-    "{% for message in messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{'<|im_start|>user\n' + message['content'] + '<|im_end|>\n'}}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{'<|im_start|>assistant\n' + message['content'] + '<|im_end|>\n' }}"\
-        "{% else %}"\
-            "{{ '<|im_start|>system\n' + message['content'] + '<|im_end|>\n' }}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ '<|im_start|>assistant\n' }}"\
+chatml_template = (
+    "{% for message in messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{'<|im_start|>user\n' + message['content'] + '<|im_end|>\n'}}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{'<|im_start|>assistant\n' + message['content'] + '<|im_end|>\n' }}"
+    "{% else %}"
+    "{{ '<|im_start|>system\n' + message['content'] + '<|im_end|>\n' }}"
     "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ '<|im_start|>assistant\n' }}"
+    "{% endif %}"
+)
 
 chatml_ollama = _ollama_template("chatml")
 
 chatml_eos_token = "<|im_end|>"
-CHAT_TEMPLATES["chatml"] = (chatml_template, chatml_eos_token, True, chatml_ollama,)
-DEFAULT_SYSTEM_MESSAGE["chatml"] = None # No system message in ChatML
+CHAT_TEMPLATES["chatml"] = (
+    chatml_template,
+    chatml_eos_token,
+    True,
+    chatml_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["chatml"] = None  # No system message in ChatML
 
 # =========================================== Mistral-1
 # Mistral Instruct doesn't allow system prompts, so we append it to the user message.
-mistral_template = \
-    "{{ bos_token }}"\
-    "{% if messages[0]['role'] == 'system' %}"\
-        "{% if messages[1]['role'] == 'user' %}"\
-            "{{ '[INST] ' + messages[0]['content'] + ' ' + messages[1]['content'] + ' [/INST]' }}"\
-            "{% set loop_messages = messages[2:] %}"\
-        "{% else %}"\
-            "{{ '[INST] ' + messages[0]['content'] + ' [/INST]' }}"\
-            "{% set loop_messages = messages[1:] %}"\
-        "{% endif %}"\
-    "{% else %}"\
-        "{% set loop_messages = messages %}"\
-    "{% endif %}"\
-    "{% for message in loop_messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{ '[INST] ' + message['content'] + ' [/INST]' }}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{ message['content'] + eos_token }}"\
-        "{% else %}"\
-            "{{ raise_exception('Only user and assistant roles are supported!') }}"\
-        "{% endif %}"\
+mistral_template = (
+    "{{ bos_token }}"
+    "{% if messages[0]['role'] == 'system' %}"
+    "{% if messages[1]['role'] == 'user' %}"
+    "{{ '[INST] ' + messages[0]['content'] + ' ' + messages[1]['content'] + ' [/INST]' }}"
+    "{% set loop_messages = messages[2:] %}"
+    "{% else %}"
+    "{{ '[INST] ' + messages[0]['content'] + ' [/INST]' }}"
+    "{% set loop_messages = messages[1:] %}"
+    "{% endif %}"
+    "{% else %}"
+    "{% set loop_messages = messages %}"
+    "{% endif %}"
+    "{% for message in loop_messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{ '[INST] ' + message['content'] + ' [/INST]' }}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{ message['content'] + eos_token }}"
+    "{% else %}"
+    "{{ raise_exception('Only user and assistant roles are supported!') }}"
+    "{% endif %}"
     "{% endfor %}"
+)
 
 # Ollama from https://www.ollama.com/library/mistral
 mistral_ollama = _ollama_template("mistral")
 
 mistral_eos_token = "eos_token"
-CHAT_TEMPLATES["mistral"] = (mistral_template, mistral_eos_token, False, mistral_ollama,)
-DEFAULT_SYSTEM_MESSAGE["mistral"] = None # No system message in Mistral
+CHAT_TEMPLATES["mistral"] = (
+    mistral_template,
+    mistral_eos_token,
+    False,
+    mistral_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["mistral"] = None  # No system message in Mistral
 
 # =========================================== Llama-2
 # Adds BOS to every convo! And weird <<SYS>> system messages.
-llama_template = \
-    "{% if messages[0]['role'] == 'system' %}"\
-        "{% if messages[1]['role'] == 'user' %}"\
-            "{{ bos_token + '[INST] <<SYS>>\n' + messages[0]['content'] + '\n<</SYS>>\n\n' + messages[1]['content'] + ' [/INST]' }}"\
-            "{% set loop_messages = messages[2:] %}"\
-        "{% else %}"\
-            "{{ bos_token + '[INST] ' + messages[0]['content'] + ' [/INST]' }}"\
-            "{% set loop_messages = messages[1:] %}"\
-        "{% endif %}"\
-    "{% else %}"\
-        "{% set loop_messages = messages %}"\
-    "{% endif %}"\
-    "{% for message in loop_messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{ bos_token + '[INST] ' + message['content'].strip() + ' [/INST]' }}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{ ' ' + message['content'].strip() + ' ' + eos_token }}"\
-        "{% else %}"\
-            "{{ raise_exception('Only user and assistant roles are supported!') }}"\
-        "{% endif %}"\
+llama_template = (
+    "{% if messages[0]['role'] == 'system' %}"
+    "{% if messages[1]['role'] == 'user' %}"
+    "{{ bos_token + '[INST] <<SYS>>\n' + messages[0]['content'] + '\n<</SYS>>\n\n' + messages[1]['content'] + ' [/INST]' }}"
+    "{% set loop_messages = messages[2:] %}"
+    "{% else %}"
+    "{{ bos_token + '[INST] ' + messages[0]['content'] + ' [/INST]' }}"
+    "{% set loop_messages = messages[1:] %}"
+    "{% endif %}"
+    "{% else %}"
+    "{% set loop_messages = messages %}"
+    "{% endif %}"
+    "{% for message in loop_messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{ bos_token + '[INST] ' + message['content'].strip() + ' [/INST]' }}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{ ' ' + message['content'].strip() + ' ' + eos_token }}"
+    "{% else %}"
+    "{{ raise_exception('Only user and assistant roles are supported!') }}"
+    "{% endif %}"
     "{% endfor %}"
+)
 
 # Ollama from https://www.ollama.com/library/llama3
 llama_ollama = _ollama_template("llama")
 
 llama_eos_token = "eos_token"
-CHAT_TEMPLATES["llama"] = (llama_template, llama_eos_token, False, llama_ollama,)
-DEFAULT_SYSTEM_MESSAGE["llama"] = None # No system message in Llama
+CHAT_TEMPLATES["llama"] = (
+    llama_template,
+    llama_eos_token,
+    False,
+    llama_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["llama"] = None  # No system message in Llama
 
 # ===========================================  Vicuna
 # https://github.com/lm-sys/FastChat/blob/main/docs/vicuna_weights_version.md#prompt-template
-vicuna_template = \
-    "{{ bos_token }}"\
-    "{% if messages[0]['role'] == 'system' %}"\
-        "{{ messages[0]['content'] + ' ' }}"\
-        "{% set loop_messages = messages[1:] %}"\
-    "{% else %}"\
-        "{{ '{system_message}' + ' ' }}"\
-        "{% set loop_messages = messages %}"\
-    "{% endif %}"\
-    "{% for message in loop_messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{ 'USER: ' + message['content'] + ' ' }}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{ 'ASSISTANT: ' + message['content'] + eos_token }}"\
-        "{% else %}"\
-            "{{ raise_exception('Only user and assistant roles are supported!') }}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ 'ASSISTANT:' }}"\
+vicuna_template = (
+    "{{ bos_token }}"
+    "{% if messages[0]['role'] == 'system' %}"
+    "{{ messages[0]['content'] + ' ' }}"
+    "{% set loop_messages = messages[1:] %}"
+    "{% else %}"
+    "{{ '{system_message}' + ' ' }}"
+    "{% set loop_messages = messages %}"
     "{% endif %}"
+    "{% for message in loop_messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{ 'USER: ' + message['content'] + ' ' }}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{ 'ASSISTANT: ' + message['content'] + eos_token }}"
+    "{% else %}"
+    "{{ raise_exception('Only user and assistant roles are supported!') }}"
+    "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ 'ASSISTANT:' }}"
+    "{% endif %}"
+)
 
 # Ollama from https://www.ollama.com/library/vicuna
 vicuna_ollama = _ollama_template("vicuna")
 
 vicuna_eos_token = "eos_token"
-CHAT_TEMPLATES["vicuna"] = (vicuna_template, vicuna_eos_token, False, vicuna_ollama,)
-DEFAULT_SYSTEM_MESSAGE["vicuna"] = "A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions."
+CHAT_TEMPLATES["vicuna"] = (
+    vicuna_template,
+    vicuna_eos_token,
+    False,
+    vicuna_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["vicuna"] = (
+    "A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions."
+)
 
 # =========================================== Vicuna Old
 # https://github.com/lm-sys/FastChat/blob/main/docs/vicuna_weights_version.md#prompt-template
-vicuna_old_template = \
-    "{{ bos_token }}"\
-    "{% if messages[0]['role'] == 'system' %}"\
-        "{{ messages[0]['content'] + '\n' }}"\
-        "{% set loop_messages = messages[1:] %}"\
-    "{% else %}"\
-        "{{ '{system_message}' + '\n' }}"\
-        "{% set loop_messages = messages %}"\
-    "{% endif %}"\
-    "{% for message in loop_messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{ '### Human: ' + message['content'] + '\n' }}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{ '### Assistant: ' + message['content'] + eos_token + '\n' }}"\
-        "{% else %}"\
-            "{{ raise_exception('Only user and assistant roles are supported!') }}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ '### Assistant:' }}"\
+vicuna_old_template = (
+    "{{ bos_token }}"
+    "{% if messages[0]['role'] == 'system' %}"
+    "{{ messages[0]['content'] + '\n' }}"
+    "{% set loop_messages = messages[1:] %}"
+    "{% else %}"
+    "{{ '{system_message}' + '\n' }}"
+    "{% set loop_messages = messages %}"
     "{% endif %}"
+    "{% for message in loop_messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{ '### Human: ' + message['content'] + '\n' }}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{ '### Assistant: ' + message['content'] + eos_token + '\n' }}"
+    "{% else %}"
+    "{{ raise_exception('Only user and assistant roles are supported!') }}"
+    "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ '### Assistant:' }}"
+    "{% endif %}"
+)
 
 vicuna_old_ollama = _ollama_template("vicuna_old")
 
 vicuna_old_eos_token = "eos_token"
-CHAT_TEMPLATES["vicuna_old"] = (vicuna_old_template, vicuna_old_eos_token, False, vicuna_old_ollama,)
-DEFAULT_SYSTEM_MESSAGE["vicuna_old"] = "A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human\\'s questions."
+CHAT_TEMPLATES["vicuna_old"] = (
+    vicuna_old_template,
+    vicuna_old_eos_token,
+    False,
+    vicuna_old_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["vicuna_old"] = (
+    "A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human\\'s questions."
+)
 
 CHAT_TEMPLATES["vicuna old"] = CHAT_TEMPLATES["vicuna_old"]
 DEFAULT_SYSTEM_MESSAGE["vicuna old"] = DEFAULT_SYSTEM_MESSAGE["vicuna_old"]
 
 # =========================================== Alpaca multi turn
 # https://github.com/tatsu-lab/stanford_alpaca Changed for multi-turn convos
-alpaca_template = \
-    "{{ bos_token }}"\
-    "{% if messages[0]['role'] == 'system' %}"\
-        "{{ messages[0]['content'] + '\n\n' }}"\
-        "{% set loop_messages = messages[1:] %}"\
-    "{% else %}"\
-        "{{ '{system_message}' + '\n\n' }}"\
-        "{% set loop_messages = messages %}"\
-    "{% endif %}"\
-    "{% for message in loop_messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{ '### Instruction:\n' + message['content'] + '\n\n' }}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{ '### Response:\n' + message['content'] + eos_token + '\n\n' }}"\
-        "{% else %}"\
-            "{{ raise_exception('Only user and assistant roles are supported!') }}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ '### Response:\n' }}"\
+alpaca_template = (
+    "{{ bos_token }}"
+    "{% if messages[0]['role'] == 'system' %}"
+    "{{ messages[0]['content'] + '\n\n' }}"
+    "{% set loop_messages = messages[1:] %}"
+    "{% else %}"
+    "{{ '{system_message}' + '\n\n' }}"
+    "{% set loop_messages = messages %}"
     "{% endif %}"
+    "{% for message in loop_messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{ '### Instruction:\n' + message['content'] + '\n\n' }}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{ '### Response:\n' + message['content'] + eos_token + '\n\n' }}"
+    "{% else %}"
+    "{{ raise_exception('Only user and assistant roles are supported!') }}"
+    "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ '### Response:\n' }}"
+    "{% endif %}"
+)
 
 alpaca_ollama = _ollama_template("alpaca")
 
 alpaca_eos_token = "eos_token"
-CHAT_TEMPLATES["alpaca"] = (alpaca_template, alpaca_eos_token, False, alpaca_ollama,)
-DEFAULT_SYSTEM_MESSAGE["alpaca"] = "Below are some instructions that describe some tasks. Write responses that appropriately complete each request."
+CHAT_TEMPLATES["alpaca"] = (
+    alpaca_template,
+    alpaca_eos_token,
+    False,
+    alpaca_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["alpaca"] = (
+    "Below are some instructions that describe some tasks. Write responses that appropriately complete each request."
+)
 
 # =========================================== Gemma
 # https://huggingface.co/google/gemma-7b-it
 # Notice we must use |trim for lstrip and rstrip. <start_of_turn> maps to 106.
 # <end_of_turn> maps to 107. user and model are normal 1 word tokens.
-gemma_template = \
-    "{{ bos_token }}"\
-    "{% if messages[0]['role'] == 'system' %}"\
-        "{{'<start_of_turn>user\n' + messages[0]['content'] | trim + ' ' + messages[1]['content'] | trim + '<end_of_turn>\n'}}"\
-        "{% set messages = messages[2:] %}"\
-    "{% endif %}"\
-    "{% for message in messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{'<start_of_turn>user\n' + message['content'] | trim + '<end_of_turn>\n'}}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{'<start_of_turn>model\n' + message['content'] | trim + '<end_of_turn>\n' }}"\
-        "{% else %}"\
-            "{{ raise_exception('Only user and assistant roles are supported!') }}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ '<start_of_turn>model\n' }}"\
+gemma_template = (
+    "{{ bos_token }}"
+    "{% if messages[0]['role'] == 'system' %}"
+    "{{'<start_of_turn>user\n' + messages[0]['content'] | trim + ' ' + messages[1]['content'] | trim + '<end_of_turn>\n'}}"
+    "{% set messages = messages[2:] %}"
     "{% endif %}"
+    "{% for message in messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{'<start_of_turn>user\n' + message['content'] | trim + '<end_of_turn>\n'}}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{'<start_of_turn>model\n' + message['content'] | trim + '<end_of_turn>\n' }}"
+    "{% else %}"
+    "{{ raise_exception('Only user and assistant roles are supported!') }}"
+    "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ '<start_of_turn>model\n' }}"
+    "{% endif %}"
+)
 
 # Ollama from https://www.ollama.com/library/gemma
 gemma_ollama = _ollama_template("gemma")
 
 gemma_eos_token = "<end_of_turn>"
-CHAT_TEMPLATES["gemma"] = (gemma_template, gemma_eos_token, True, gemma_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gemma"] = None # No system message in Gemma
+CHAT_TEMPLATES["gemma"] = (
+    gemma_template,
+    gemma_eos_token,
+    True,
+    gemma_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gemma"] = None  # No system message in Gemma
 
 # =========================================== Gemma with ChatML instead
 # We find using <eos> is still more appropriate!
@@ -315,11 +376,16 @@ gemma_chatml_template = "{{ bos_token }}" + chatml_template
 gemma_chatml_ollama = _ollama_template("gemma_chatml")
 
 gemma_chatml_eos_token = (
-    {"<start_of_turn>" : "<|im_start|>", "<eos>" : "<|im_end|>"},
+    {"<start_of_turn>": "<|im_start|>", "<eos>": "<|im_end|>"},
     "<|im_end|>",
 )
-CHAT_TEMPLATES["gemma_chatml"] = (gemma_chatml_template, gemma_chatml_eos_token, True, gemma_chatml_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gemma_chatml"] = None # No system message in Gemma
+CHAT_TEMPLATES["gemma_chatml"] = (
+    gemma_chatml_template,
+    gemma_chatml_eos_token,
+    True,
+    gemma_chatml_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gemma_chatml"] = None  # No system message in Gemma
 
 # =========================================== Gemma 2
 # Same as Gemma 1, but with sliding window attention!
@@ -327,73 +393,100 @@ DEFAULT_SYSTEM_MESSAGE["gemma_chatml"] = None # No system message in Gemma
 gemma2_template = gemma_template
 gemma2_ollama = _ollama_template("gemma2")
 gemma2_eos_token = "<end_of_turn>"
-CHAT_TEMPLATES["gemma2"] = (gemma2_template, gemma2_eos_token, True, gemma2_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gemma2"] = None # No system message in Gemma 2
+CHAT_TEMPLATES["gemma2"] = (
+    gemma2_template,
+    gemma2_eos_token,
+    True,
+    gemma2_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gemma2"] = None  # No system message in Gemma 2
 
 # =========================================== Gemma 2 with ChatML instead
 gemma2_chatml_template = gemma_chatml_template
 gemma2_chatml_ollama = _ollama_template("gemma2_chatml")
 gemma2_chatml_eos_token = gemma_chatml_eos_token
-CHAT_TEMPLATES["gemma2_chatml"] = (gemma2_chatml_template, gemma2_chatml_eos_token, True, gemma2_chatml_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gemma2_chatml"] = None # No system message in Gemma 2
+CHAT_TEMPLATES["gemma2_chatml"] = (
+    gemma2_chatml_template,
+    gemma2_chatml_eos_token,
+    True,
+    gemma2_chatml_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gemma2_chatml"] = None  # No system message in Gemma 2
 
 # =========================================== Llama-3
 # Weirdly \n\n is needed?
-llama3_template = \
-    "{{ bos_token }}"\
-    "{% for message in messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{ '<|start_header_id|>user<|end_header_id|>\n\n' + message['content'] | trim + '<|eot_id|>' }}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' + message['content'] | trim + '<|eot_id|>' }}"\
-        "{% else %}"\
-            "{{ '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n' + message['content'] | trim + '<|eot_id|>' }}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' }}"\
+llama3_template = (
+    "{{ bos_token }}"
+    "{% for message in messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{ '<|start_header_id|>user<|end_header_id|>\n\n' + message['content'] | trim + '<|eot_id|>' }}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' + message['content'] | trim + '<|eot_id|>' }}"
+    "{% else %}"
+    "{{ '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n' + message['content'] | trim + '<|eot_id|>' }}"
     "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' }}"
+    "{% endif %}"
+)
 
 # Ollama from https://www.ollama.com/library/llama3
 llama3_ollama = _ollama_template("llama-3")
 
 llama3_template_eos_token = "eos_token"
 
-CHAT_TEMPLATES["llama-3"] = (llama3_template, llama3_template_eos_token, False, llama3_ollama,)
-DEFAULT_SYSTEM_MESSAGE["llama-3"] = None # No system message in Llama-3
+CHAT_TEMPLATES["llama-3"] = (
+    llama3_template,
+    llama3_template_eos_token,
+    False,
+    llama3_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["llama-3"] = None  # No system message in Llama-3
 
-CHAT_TEMPLATES["llama3"] = (llama3_template, llama3_template_eos_token, False, llama3_ollama,)
-DEFAULT_SYSTEM_MESSAGE["llama3"] = None # No system message in Llama-3
+CHAT_TEMPLATES["llama3"] = (
+    llama3_template,
+    llama3_template_eos_token,
+    False,
+    llama3_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["llama3"] = None  # No system message in Llama-3
 
 
 # =========================================== Phi-3
 # "{{ bos_token }}"\ # Phi-3.5 removes BOS?
-phi3_template = \
-    "{% for message in messages %}"\
-        "{% if message['role'] == 'user' %}"\
-            "{{'<|user|>\n' + message['content'] + '<|end|>\n'}}"\
-        "{% elif message['role'] == 'assistant' %}"\
-            "{{'<|assistant|>\n' + message['content'] + '<|end|>\n'}}"\
-        "{% else %}"\
-            "{{'<|' + message['role'] + '|>\n' + message['content'] + '<|end|>\n'}}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ '<|assistant|>\n' }}"\
+phi3_template = (
+    "{% for message in messages %}"
+    "{% if message['role'] == 'user' %}"
+    "{{'<|user|>\n' + message['content'] + '<|end|>\n'}}"
+    "{% elif message['role'] == 'assistant' %}"
+    "{{'<|assistant|>\n' + message['content'] + '<|end|>\n'}}"
+    "{% else %}"
+    "{{'<|' + message['role'] + '|>\n' + message['content'] + '<|end|>\n'}}"
     "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ '<|assistant|>\n' }}"
+    "{% endif %}"
+)
 
 # Ollama from https://www.ollama.com/library/phi3
 phi3_ollama = _ollama_template("phi-3")
 
 phi3_template_eos_token = "<|end|>"
-CHAT_TEMPLATES["phi-3"]   = (phi3_template, phi3_template_eos_token, False, phi3_ollama,)
-DEFAULT_SYSTEM_MESSAGE["phi-3"] = None # No system message in Phi-3
+CHAT_TEMPLATES["phi-3"] = (
+    phi3_template,
+    phi3_template_eos_token,
+    False,
+    phi3_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["phi-3"] = None  # No system message in Phi-3
 
-CHAT_TEMPLATES["phi-35"]  = CHAT_TEMPLATES["phi-3"]
-DEFAULT_SYSTEM_MESSAGE["phi-35"] = None # No system message in Phi-3.5
+CHAT_TEMPLATES["phi-35"] = CHAT_TEMPLATES["phi-3"]
+DEFAULT_SYSTEM_MESSAGE["phi-35"] = None  # No system message in Phi-3.5
 
 CHAT_TEMPLATES["phi-3.5"] = CHAT_TEMPLATES["phi-3"]
-DEFAULT_SYSTEM_MESSAGE["phi-3.5"] = None # No system message in Phi-3.5
+DEFAULT_SYSTEM_MESSAGE["phi-3.5"] = None  # No system message in Phi-3.5
 
 # =========================================== Llama-3.1
 """
@@ -412,8 +505,7 @@ tokenizer.apply_chat_template(
 )
 """
 
-llama31_template = \
-"""{{- bos_token }}
+llama31_template = """{{- bos_token }}
 {%- if custom_tools is defined %}
     {%- set tools = custom_tools %}
 {%- endif %}
@@ -528,11 +620,25 @@ llama31_template = \
 llama31_ollama = _ollama_template("llama-3.1")
 
 llama31_template_eos_token = "eos_token"
-CHAT_TEMPLATES["llama-3.1"] = (llama31_template, llama31_template_eos_token, False, llama31_ollama,)
-DEFAULT_SYSTEM_MESSAGE["llama-3.1"] = "" # Llama3.1 default system message is empty + the dates
+CHAT_TEMPLATES["llama-3.1"] = (
+    llama31_template,
+    llama31_template_eos_token,
+    False,
+    llama31_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["llama-3.1"] = (
+    ""  # Llama3.1 default system message is empty + the dates
+)
 
-CHAT_TEMPLATES["llama-31"]  = (llama31_template, llama31_template_eos_token, False, llama31_ollama,)
-DEFAULT_SYSTEM_MESSAGE["llama-31"] = "" # Llama3.1 default system message is empty + the dates
+CHAT_TEMPLATES["llama-31"] = (
+    llama31_template,
+    llama31_template_eos_token,
+    False,
+    llama31_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["llama-31"] = (
+    ""  # Llama3.1 default system message is empty + the dates
+)
 
 for version in ("llama-3.2", "llama-3.3", "llama-32", "llama-33"):
     CHAT_TEMPLATES[version] = CHAT_TEMPLATES["llama-3.1"]
@@ -540,8 +646,7 @@ for version in ("llama-3.2", "llama-3.3", "llama-32", "llama-33"):
 
 
 # =========================================== Qwen 2.5
-qwen25_template = \
-"""{%- if tools %}
+qwen25_template = """{%- if tools %}
     {{- \'<|im_start|>system\\n\' }}
     {%- if messages[0][\'role\'] == \'system\' %}
         {{- messages[0][\'content\'] }}
@@ -596,53 +701,89 @@ qwen25_template = \
 qwen25_ollama = _ollama_template("qwen-2.5")
 
 qwen25_template_eos_token = "eos_token"
-qwen25_default_system_message = "You are Qwen, created by Alibaba Cloud. You are a helpful assistant."
-CHAT_TEMPLATES["qwen-2.5"] = (qwen25_template, qwen25_template_eos_token, False, qwen25_ollama,)
-DEFAULT_SYSTEM_MESSAGE["qwen-2.5"] = qwen25_default_system_message # No system message in Qwen 2.5
+qwen25_default_system_message = (
+    "You are Qwen, created by Alibaba Cloud. You are a helpful assistant."
+)
+CHAT_TEMPLATES["qwen-2.5"] = (
+    qwen25_template,
+    qwen25_template_eos_token,
+    False,
+    qwen25_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["qwen-2.5"] = (
+    qwen25_default_system_message  # No system message in Qwen 2.5
+)
 
-CHAT_TEMPLATES["qwen-25"]  = (qwen25_template, qwen25_template_eos_token, False, qwen25_ollama,)
-DEFAULT_SYSTEM_MESSAGE["qwen-25"] = qwen25_default_system_message # No system message in Qwen 2.5
+CHAT_TEMPLATES["qwen-25"] = (
+    qwen25_template,
+    qwen25_template_eos_token,
+    False,
+    qwen25_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["qwen-25"] = (
+    qwen25_default_system_message  # No system message in Qwen 2.5
+)
 
-CHAT_TEMPLATES["qwen25"]   = (qwen25_template, qwen25_template_eos_token, False, qwen25_ollama,)
-DEFAULT_SYSTEM_MESSAGE["qwen25"] = qwen25_default_system_message # No system message in Qwen 2.5
+CHAT_TEMPLATES["qwen25"] = (
+    qwen25_template,
+    qwen25_template_eos_token,
+    False,
+    qwen25_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["qwen25"] = (
+    qwen25_default_system_message  # No system message in Qwen 2.5
+)
 
-CHAT_TEMPLATES["qwen2.5"]  = (qwen25_template, qwen25_template_eos_token, False, qwen25_ollama,)
-DEFAULT_SYSTEM_MESSAGE["qwen2.5"] = qwen25_default_system_message # No system message in Qwen 2.5
+CHAT_TEMPLATES["qwen2.5"] = (
+    qwen25_template,
+    qwen25_template_eos_token,
+    False,
+    qwen25_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["qwen2.5"] = (
+    qwen25_default_system_message  # No system message in Qwen 2.5
+)
 
 # =========================================== Phi-4
 # "{{ bos_token }}"\ # Phi-4 removes BOS?
-phi4_template = \
-    "{% for message in messages %}"\
-        "{% if (message['role'] == 'system') %}"\
-            "{{'<|im_start|>system<|im_sep|>' + message['content'] + '<|im_end|>'}}"\
-        "{% elif (message['role'] == 'user') %}"\
-            "{{'<|im_start|>user<|im_sep|>' + message['content'] + '<|im_end|>'}}"\
-        "{% elif (message['role'] == 'assistant') %}"\
-            "{{'<|im_start|>assistant<|im_sep|>' + message['content'] + '<|im_end|>'}}"\
-        "{% endif %}"\
-    "{% endfor %}"\
-    "{% if add_generation_prompt %}"\
-        "{{ '<|im_start|>assistant<|im_sep|>' }}"\
+phi4_template = (
+    "{% for message in messages %}"
+    "{% if (message['role'] == 'system') %}"
+    "{{'<|im_start|>system<|im_sep|>' + message['content'] + '<|im_end|>'}}"
+    "{% elif (message['role'] == 'user') %}"
+    "{{'<|im_start|>user<|im_sep|>' + message['content'] + '<|im_end|>'}}"
+    "{% elif (message['role'] == 'assistant') %}"
+    "{{'<|im_start|>assistant<|im_sep|>' + message['content'] + '<|im_end|>'}}"
     "{% endif %}"
+    "{% endfor %}"
+    "{% if add_generation_prompt %}"
+    "{{ '<|im_start|>assistant<|im_sep|>' }}"
+    "{% endif %}"
+)
 
-_phi4_ollama_template = \
-    "{{ if .System }}<|im_start|><|system|><|im_sep|>{{ .System }}<|im_end|>{{ end }}"\
-    "{{ if .Prompt }}<|im_start|><|user|><|im_sep|>{{ .Prompt }}<|im_end|>{{ end }}"\
+_phi4_ollama_template = (
+    "{{ if .System }}<|im_start|><|system|><|im_sep|>{{ .System }}<|im_end|>{{ end }}"
+    "{{ if .Prompt }}<|im_start|><|user|><|im_sep|>{{ .Prompt }}<|im_end|>{{ end }}"
     "<|im_start|><|assistant|><|im_sep|>{{ .Response }}<|im_end|>"
+)
 
 # Ollama from https://www.ollama.com/library/phi4 is different
 phi4_ollama = _ollama_template("phi-4")
 
 phi4_template_eos_token = "<|im_end|>"
-CHAT_TEMPLATES["phi-4"] = (phi4_template, phi4_template_eos_token, False, phi4_ollama,)
-DEFAULT_SYSTEM_MESSAGE["phi-4"] = None # No system message in Phi-4
+CHAT_TEMPLATES["phi-4"] = (
+    phi4_template,
+    phi4_template_eos_token,
+    False,
+    phi4_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["phi-4"] = None  # No system message in Phi-4
 
 
 # =========================================== Gemma-3
 # Obtained via
 # print(tokenizer.chat_template.replace("}\n", "####").replace("\n", "\\n").replace("####", "}\n"))
-gemma3_template = \
-"""{{ bos_token }}
+gemma3_template = """{{ bos_token }}
 {%- if messages[0]['role'] == 'system' -%}
     {%- if messages[0]['content'] is string -%}
         {%- set first_user_prefix = messages[0]['content'] + '\n\n' -%}
@@ -688,16 +829,25 @@ gemma3_template = \
 gemma3_ollama = _ollama_template("gemma-3")
 
 gemma3_template_eos_token = "<end_of_turn>"
-CHAT_TEMPLATES["gemma-3"] = (gemma3_template, gemma3_template_eos_token, False, gemma3_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gemma-3"] = None # No system message in Gemma-3
+CHAT_TEMPLATES["gemma-3"] = (
+    gemma3_template,
+    gemma3_template_eos_token,
+    False,
+    gemma3_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gemma-3"] = None  # No system message in Gemma-3
 
-CHAT_TEMPLATES["gemma3"] = (gemma3_template, gemma3_template_eos_token, False, gemma3_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gemma3"] = None # No system message in Gemma-3
+CHAT_TEMPLATES["gemma3"] = (
+    gemma3_template,
+    gemma3_template_eos_token,
+    False,
+    gemma3_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gemma3"] = None  # No system message in Gemma-3
 
 # =========================================== Qwen-3
 # Official Qwen-3 chat template (see https://ollama.com/library/qwen3/blobs/eb4402837c78)
-qwen3_template = \
-"""
+qwen3_template = """
 {%- if tools %}
     {{- '<|im_start|>system\n' }}
     {%- if messages[0].role == 'system' %}
@@ -800,17 +950,26 @@ qwen3_template = \
 
 qwen3_ollama = _ollama_template("qwen-3")
 qwen3_template_eos_token = "<|im_end|>"
-CHAT_TEMPLATES["qwen-3"] = (qwen3_template, qwen3_template_eos_token, False, qwen3_ollama,)
-DEFAULT_SYSTEM_MESSAGE["qwen-3"] = None # No default system message for Qwen-3
+CHAT_TEMPLATES["qwen-3"] = (
+    qwen3_template,
+    qwen3_template_eos_token,
+    False,
+    qwen3_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["qwen-3"] = None  # No default system message for Qwen-3
 
-CHAT_TEMPLATES["qwen3"] = (qwen3_template, qwen3_template_eos_token, False, qwen3_ollama,)
-DEFAULT_SYSTEM_MESSAGE["qwen3"] = None # No default system message for Qwen-3
+CHAT_TEMPLATES["qwen3"] = (
+    qwen3_template,
+    qwen3_template_eos_token,
+    False,
+    qwen3_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["qwen3"] = None  # No default system message for Qwen-3
 
 # =========================================== Gemma-3n
 # Obtained via
 # print(tokenizer.chat_template.replace("}\n", "####").replace("\n", "\\n").replace("####", "}\n"))
-gemma3n_template = \
-"""{{ bos_token }}
+gemma3n_template = """{{ bos_token }}
 {%- if messages[0]['role'] == 'system' -%}
     {%- if messages[0]['content'] is string -%}
         {%- set first_user_prefix = messages[0]['content'] + '\n\n' -%}
@@ -857,17 +1016,26 @@ gemma3n_template = \
 # Ollama from https://ollama.com/library/gemma3n/blobs/e0a42594d802
 gemma3n_ollama = _ollama_template("gemma-3n")
 gemma3n_template_eos_token = "<end_of_turn>"
-CHAT_TEMPLATES["gemma-3n"] = (gemma3n_template, gemma3n_template_eos_token, False, gemma3n_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gemma-3n"] = None # No system message in Gemma-3n
+CHAT_TEMPLATES["gemma-3n"] = (
+    gemma3n_template,
+    gemma3n_template_eos_token,
+    False,
+    gemma3n_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gemma-3n"] = None  # No system message in Gemma-3n
 
-CHAT_TEMPLATES["gemma3n"] = (gemma3n_template, gemma3n_template_eos_token, False, gemma3n_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gemma3n"] = None # No system message in Gemma-3n
+CHAT_TEMPLATES["gemma3n"] = (
+    gemma3n_template,
+    gemma3n_template_eos_token,
+    False,
+    gemma3n_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gemma3n"] = None  # No system message in Gemma-3n
 
 # =========================================== GPT-OSS
 # Obtained via
 # print(tokenizer.chat_template.replace("}\n", "####").replace("\n", "\\n").replace("####", "}\n"))
-gptoss_template = \
-"""{#-
+gptoss_template = """{#-
   In addition to the normal inputs of `messages` and `tools`, this template also accepts the
   following kwargs:
   - "builtin_tools": A list, can contain "browser" and/or "python".
@@ -1217,8 +1385,7 @@ gptoss_template = \
 {%- endif -%}"""
 
 # Ollama from https://ollama.com/library/gpt-oss
-gptoss_ollama = \
-'''
+gptoss_ollama = '''
 FROM {__FILE_LOCATION__}
 TEMPLATE """<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
 Knowledge cutoff: 2024-06
@@ -1399,15 +1566,24 @@ PARAMETER top_p 1.0
 '''
 
 gptoss_template_template_eos_token = "<|return|>"
-CHAT_TEMPLATES["gpt-oss"] = (gptoss_template, gptoss_template_template_eos_token, False, gptoss_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gpt-oss"] = None # No system message in GPT-oss
+CHAT_TEMPLATES["gpt-oss"] = (
+    gptoss_template,
+    gptoss_template_template_eos_token,
+    False,
+    gptoss_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gpt-oss"] = None  # No system message in GPT-oss
 
-CHAT_TEMPLATES["gptoss"] = (gptoss_template, gptoss_template_template_eos_token, False, gptoss_ollama,)
-DEFAULT_SYSTEM_MESSAGE["gptoss"] = None # No system message in GPT-oss
+CHAT_TEMPLATES["gptoss"] = (
+    gptoss_template,
+    gptoss_template_template_eos_token,
+    False,
+    gptoss_ollama,
+)
+DEFAULT_SYSTEM_MESSAGE["gptoss"] = None  # No system message in GPT-oss
 
 # =========================================== Qwen3-Instruct
-qwen3_instruct_template = \
-'''{%- if tools %}
+qwen3_instruct_template = """{%- if tools %}
     {{- '<|im_start|>system\\n' }}
     {%- if messages[0].role == 'system' %}
         {{- messages[0].content + '\\n\\n' }}
@@ -1492,16 +1668,20 @@ qwen3_instruct_template = \
 {%- endfor %}
 {%- if add_generation_prompt %}
     {{- '<|im_start|>assistant\\n' }}
-{%- endif %}'''
+{%- endif %}"""
 
 qwen3_template_eos_token = "<|im_end|>"
-CHAT_TEMPLATES["qwen3-instruct"] = (qwen3_instruct_template, qwen3_template_eos_token, False, _ollama_template("qwen3-instruct"),)
-DEFAULT_SYSTEM_MESSAGE["qwen3-instruct"] = None # No system message in Qwen3
+CHAT_TEMPLATES["qwen3-instruct"] = (
+    qwen3_instruct_template,
+    qwen3_template_eos_token,
+    False,
+    _ollama_template("qwen3-instruct"),
+)
+DEFAULT_SYSTEM_MESSAGE["qwen3-instruct"] = None  # No system message in Qwen3
 
 
 # =========================================== Qwen3-Thinking
-qwen3_thinking_template = \
-'''{%- if tools %}
+qwen3_thinking_template = """{%- if tools %}
     {{- '<|im_start|>system\\n' }}
     {%- if messages[0].role == 'system' %}
         {{- messages[0].content + '\\n\\n' }}
@@ -1586,7 +1766,7 @@ qwen3_thinking_template = \
 {%- endfor %}
 {%- if add_generation_prompt %}
     {{- '<|im_start|>assistant\n<think>\n' }}
-{%- endif %}'''
+{%- endif %}"""
 
 CHAT_TEMPLATES["qwen3-thinking"] = (
     qwen3_thinking_template,
@@ -1594,26 +1774,29 @@ CHAT_TEMPLATES["qwen3-thinking"] = (
     False,
     _ollama_template("qwen3-thinking"),
 )
-DEFAULT_SYSTEM_MESSAGE["qwen3-thinking"] = None # No system message in Qwen3
+DEFAULT_SYSTEM_MESSAGE["qwen3-thinking"] = None  # No system message in Qwen3
 
 
 # =========================================== Liquid-LFM2
-liquid_lfm2_template = \
-'''
+liquid_lfm2_template = """
 {{bos_token}}{% for message in messages %}{{'<|im_start|>' + message['role'] + '
 ' + message['content'] + '<|im_end|>' + '
 '}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
-' }}{% endif %}'''
+' }}{% endif %}"""
 
 liquid_lfm2_template_eos_token = "<|im_end|>"
-CHAT_TEMPLATES["lfm-2"] = (liquid_lfm2_template, liquid_lfm2_template_eos_token, False, None)
-DEFAULT_SYSTEM_MESSAGE["lfm-2"] = None # No system message in Phi-3
+CHAT_TEMPLATES["lfm-2"] = (
+    liquid_lfm2_template,
+    liquid_lfm2_template_eos_token,
+    False,
+    None,
+)
+DEFAULT_SYSTEM_MESSAGE["lfm-2"] = None  # No system message in Phi-3
 
 
 # =========================================== Starling-LM
 
-starling_template = \
-"""{{ bos_token }}
+starling_template = """{{ bos_token }}
 {%- for message in messages %}
     {{ 'GPT4 Correct ' + message['role'].title() + ': ' + message['content'] + '<|end_of_turn|>' }}
 {%- endfor %}
@@ -1625,14 +1808,18 @@ starling_template = \
 starling_ollama = _ollama_template("starling")
 
 starling_template_eos_token = "<|end_of_turn|>"
-CHAT_TEMPLATES["starling"] = (starling_template, starling_template_eos_token, False, starling_ollama)
+CHAT_TEMPLATES["starling"] = (
+    starling_template,
+    starling_template_eos_token,
+    False,
+    starling_ollama,
+)
 DEFAULT_SYSTEM_MESSAGE["starling"] = None
 
 
 # =========================================== Yi-chat
 
-yi_chat_template = \
-"""
+yi_chat_template = """
 {% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '
 ' + message['content'] + '<|im_end|>' + '
 '}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
@@ -1643,10 +1830,18 @@ yi_chat_template = \
 yi_chat_ollama = _ollama_template("yi-chat")
 
 yi_chat_template_eos_token = "<|endoftext|>"
-CHAT_TEMPLATES["yi-chat"] = (yi_chat_template, yi_chat_template_eos_token, False, yi_chat_ollama)
+CHAT_TEMPLATES["yi-chat"] = (
+    yi_chat_template,
+    yi_chat_template_eos_token,
+    False,
+    yi_chat_ollama,
+)
 DEFAULT_SYSTEM_MESSAGE["yi-chat"] = None
 
-def _change_system_message(template: str, type_chat_template: str, system_message: str = None):
+
+def _change_system_message(
+    template: str, type_chat_template: str, system_message: str = None
+):
     system_message_pattern = r"\{system_message\}"
 
     # For predefined templates, check if default system message exists
@@ -1666,14 +1861,18 @@ def _change_system_message(template: str, type_chat_template: str, system_messag
 
         if has_placeholder:
             if system_message is None:
-                raise ValueError("Unsloth: You need to provide a system message for custom templates.")
+                raise ValueError(
+                    "Unsloth: You need to provide a system message for custom templates."
+                )
             new_template = re.sub(system_message_pattern, system_message, template)
             return new_template, system_message
 
         return template, system_message
 
     # For predefined templates with default system message
-    message_to_use = system_message if system_message is not None else default_system_message
+    message_to_use = (
+        system_message if system_message is not None else default_system_message
+    )
     new_template = re.sub(system_message_pattern, message_to_use, template)
 
     return new_template, message_to_use
@@ -1682,16 +1881,22 @@ def _change_system_message(template: str, type_chat_template: str, system_messag
 def get_chat_template(
     tokenizer,
     chat_template = "chatml",
-    mapping = {"role" : "role", "content" : "content", "user" : "user", "assistant" : "assistant"},
+    mapping = {
+        "role": "role",
+        "content": "content",
+        "user": "user",
+        "assistant": "assistant",
+    },
     map_eos_token = True,
     system_message = None,
 ):
-    assert(type(map_eos_token) is bool)
+    assert type(map_eos_token) is bool
     old_tokenizer = tokenizer
 
     IS_GEMMA = False
     if tokenizer.__class__.__name__.startswith("Gemma"):
-        if chat_template == "chatml": chat_template = "gemma_chatml"
+        if chat_template == "chatml":
+            chat_template = "gemma_chatml"
         IS_GEMMA = True
 
     # We add a check for Llama-3
@@ -1712,32 +1917,42 @@ def get_chat_template(
     same_padding_token = False
     type_chat_template = None
 
-    if type(chat_template) in (list, tuple,):
+    if type(chat_template) in (
+        list,
+        tuple,
+    ):
         # For changing system message later
         # Since it's not supported yet, we will raise an error first!
         type_chat_template = chat_template[0].lower()
         chat_template, stop_word = chat_template
-        assert(type(chat_template) is str)
-        assert(type(stop_word) is str)
+        assert type(chat_template) is str
+        assert type(stop_word) is str
         ollama_modelfile = None
 
     elif type(chat_template) is str:
         # For changing system message later
         type_chat_template = chat_template.lower()
 
-        chat_template, stop_word, yes_map_eos_token, ollama_modelfile = CHAT_TEMPLATES[chat_template]
+        chat_template, stop_word, yes_map_eos_token, ollama_modelfile = CHAT_TEMPLATES[
+            chat_template
+        ]
 
         # Check mapping to eos_token
-        if not map_eos_token and yes_map_eos_token: map_eos_token = True
-        if not yes_map_eos_token and map_eos_token: map_eos_token = False
+        if not map_eos_token and yes_map_eos_token:
+            map_eos_token = True
+        if not yes_map_eos_token and map_eos_token:
+            map_eos_token = False
 
-        if type(stop_word) in (list, tuple,):
+        if type(stop_word) in (
+            list,
+            tuple,
+        ):
             token_mapping, stop_word = stop_word
-            assert(type(token_mapping) is dict)
+            assert type(token_mapping) is dict
         else:
             token_mapping = None
 
-        assert(type(stop_word) is str)
+        assert type(stop_word) is str
 
         # Check fast tokenizer
         if not is_fast_tokenizer:
@@ -1764,13 +1979,17 @@ def get_chat_template(
                 elif old_count == 0:
                     raise RuntimeError(f"{old_token} was not part of the tokenizer!")
                 else:
-                    string_vocab = string_vocab.replace(f'"{old_token}"', f'"{new_token}"')
+                    string_vocab = string_vocab.replace(
+                        f'"{old_token}"', f'"{new_token}"'
+                    )
                 pass
             pass
 
-            if map_eos_token and (not stop_word in token_mapping.values()):
+            if map_eos_token and (stop_word not in token_mapping.values()):
                 # Do not map 107 = <|im_end|> and 1 = <|im_end|>. This will reduce the vocab size by 1
-                logger.warning_once(f"Unsloth: Will map {stop_word} to EOS = {tokenizer.eos_token}.")
+                logger.warning_once(
+                    f"Unsloth: Will map {stop_word} to EOS = {tokenizer.eos_token}."
+                )
                 string_vocab = string_vocab.replace(tokenizer.eos_token, stop_word)
             pass
 
@@ -1798,12 +2017,18 @@ def get_chat_template(
                 pass
 
                 # Must fix the sentence piece tokenizer since there's no tokenizer.model file!
-                tokenizer = fix_sentencepiece_tokenizer(tokenizer, new_tokenizer, token_mapping,)
+                tokenizer = fix_sentencepiece_tokenizer(
+                    tokenizer,
+                    new_tokenizer,
+                    token_mapping,
+                )
             else:
                 pass
 
         elif map_eos_token and (stop_word != "eos_token"):
-            logger.warning_once(f"Unsloth: Will map {stop_word} to EOS = {tokenizer.eos_token}.")
+            logger.warning_once(
+                f"Unsloth: Will map {stop_word} to EOS = {tokenizer.eos_token}."
+            )
 
             # Replaces the old EOS token with a new one.
             # Useful for ChatML <|im_end|> for example.
@@ -1845,43 +2070,60 @@ def get_chat_template(
             )
 
             # Must fix the sentence piece tokenizer since there's no tokenizer.model file!
-            token_mapping = { old_eos_token : stop_word, }
-            tokenizer = fix_sentencepiece_tokenizer(tokenizer, new_tokenizer, token_mapping,)
+            token_mapping = {
+                old_eos_token: stop_word,
+            }
+            tokenizer = fix_sentencepiece_tokenizer(
+                tokenizer,
+                new_tokenizer,
+                token_mapping,
+            )
         pass
 
     else:
         raise TypeError(
-            f"Unsloth: `chat_template` must be a tuple of (your_template, eos_token,) or one of\n"\
+            f"Unsloth: `chat_template` must be a tuple of (your_template, eos_token,) or one of\n"
             f"{CHAT_TEMPLATES.keys()}"
         )
 
     # Careful on Gemma
     # bos_token is a must or else losses become too high
-    if IS_GEMMA and not chat_template.startswith(("{{ bos_token }}", "{{- bos_token }}")):
+    if IS_GEMMA and not chat_template.startswith(
+        ("{{ bos_token }}", "{{- bos_token }}")
+    ):
         chat_template = "{{ bos_token }}" + chat_template
 
     # For ShareGPT role -> from and content -> value
-    new_chat_template = chat_template\
-        .replace("'role'",      "'" + mapping["role"]      + "'")\
-        .replace("'content'",   "'" + mapping["content"]   + "'")\
-        .replace("'user'",      "'" + mapping["user"]      + "'")\
+    new_chat_template = (
+        chat_template.replace("'role'", "'" + mapping["role"] + "'")
+        .replace("'content'", "'" + mapping["content"] + "'")
+        .replace("'user'", "'" + mapping["user"] + "'")
         .replace("'assistant'", "'" + mapping["assistant"] + "'")
+    )
 
     _, tokenizer = patch_tokenizer(model = None, tokenizer = tokenizer)
     tokenizer.padding_side = old_padding_side
 
     # If not normal HF, we add a check to make old templates work
-    if mapping != {"role" : "role", "content" : "content", "user" : "user", "assistant" : "assistant"}:
-        chat_template = \
-            "{% if 'role' in messages[0] %}" + \
-            chat_template + \
-            "{% else %}" + \
-            new_chat_template + \
-            "{% endif %}"
+    if mapping != {
+        "role": "role",
+        "content": "content",
+        "user": "user",
+        "assistant": "assistant",
+    }:
+        chat_template = (
+            "{% if 'role' in messages[0] %}"
+            + chat_template
+            + "{% else %}"
+            + new_chat_template
+            + "{% endif %}"
+        )
     else:
         chat_template = new_chat_template
 
-    chat_template, system_message = _change_system_message(chat_template, type_chat_template, system_message)
+    chat_template, system_message = _change_system_message(
+        chat_template, type_chat_template, system_message
+    )
 
     tokenizer.chat_template = chat_template
 
@@ -1889,13 +2131,16 @@ def get_chat_template(
     old_pad_token = getattr(old_tokenizer, "pad_token", None)
     old_bos_token = getattr(old_tokenizer, "bos_token", None)
     old_unk_token = getattr(old_tokenizer, "unk_token", None)
-    new_pad_token = getattr(tokenizer,     "pad_token", None)
-    new_bos_token = getattr(tokenizer,     "bos_token", None)
-    new_unk_token = getattr(tokenizer,     "unk_token", None)
-    if old_bos_token != new_bos_token: tokenizer.bos_token = old_bos_token
-    if old_unk_token != new_unk_token: tokenizer.unk_token = old_unk_token
+    new_pad_token = getattr(tokenizer, "pad_token", None)
+    new_bos_token = getattr(tokenizer, "bos_token", None)
+    new_unk_token = getattr(tokenizer, "unk_token", None)
+    if old_bos_token != new_bos_token:
+        tokenizer.bos_token = old_bos_token
+    if old_unk_token != new_unk_token:
+        tokenizer.unk_token = old_unk_token
     if not same_padding_token:
-        if old_pad_token != new_pad_token: tokenizer.pad_token = old_pad_token
+        if old_pad_token != new_pad_token:
+            tokenizer.pad_token = old_pad_token
 
     # stopping_criteria = create_stopping_criteria(tokenizer, stop_word)
 
@@ -1904,14 +2149,14 @@ def get_chat_template(
 
     # Add Ollama
     tokenizer._ollama_modelfile = ollama_modelfile
-    tokenizer._system_message   = system_message
-    return tokenizer#, stopping_criteria
+    tokenizer._system_message = system_message
+    return tokenizer  # , stopping_criteria
 
 
 def remove_special_tokens(tokenizer, prompt):
     # Removes double BOS token
     if prompt.startswith(tokenizer.bos_token):
-        prompt = prompt[len(tokenizer.bos_token):]
+        prompt = prompt[len(tokenizer.bos_token) :]
     return prompt
 
 
@@ -1922,12 +2167,14 @@ def _parse_combined_prompt(combined_prompt, dataset):
     for column in possible_columns:
         if column not in dataset_columns:
             raise KeyError(
-                f"Unsloth: Your prompt includes '{column}' but this does not exist in the dataset. "\
+                f"Unsloth: Your prompt includes '{column}' but this does not exist in the dataset. "
                 f"Only allowed columns are {list(dataset_columns)}"
             )
 
     # Find [[...]]
-    optional_prompts = list(re.finditer(r"\[\[.+?\]\]", combined_prompt, flags = re.DOTALL | re.MULTILINE))
+    optional_prompts = list(
+        re.finditer(r"\[\[.+?\]\]", combined_prompt, flags = re.DOTALL | re.MULTILINE)
+    )
     optional_prompts = [(x.span(), x.group(0)) for x in optional_prompts]
 
     final_optional_prompts = []
@@ -1935,25 +2182,30 @@ def _parse_combined_prompt(combined_prompt, dataset):
         # Add left
         left = optional_prompts[0]
         l = left[0][0]
-        if l != 0: final_optional_prompts.append(combined_prompt[:l])
+        if l != 0:
+            final_optional_prompts.append(combined_prompt[:l])
 
         # Add in between
         for left, right in zip(optional_prompts[:-1], optional_prompts[1:]):
             l, r = left[0][-1], right[0][0]
             final_optional_prompts.append(left)
-            if l != r: final_optional_prompts.append(combined_prompt[l : r])
+            if l != r:
+                final_optional_prompts.append(combined_prompt[l:r])
         final_optional_prompts.append(optional_prompts[-1])
 
         # Add right
         right = optional_prompts[-1]
         r = right[0][1]
-        if r != len(combined_prompt): final_optional_prompts.append(combined_prompt[r:])
+        if r != len(combined_prompt):
+            final_optional_prompts.append(combined_prompt[r:])
     else:
         # Just add in the entire string
         final_optional_prompts.append(combined_prompt)
 
-    check_combined = "".join(x if type(x) is str else x[1] for x in final_optional_prompts)
-    assert(combined_prompt == check_combined)
+    check_combined = "".join(
+        x if type(x) is str else x[1] for x in final_optional_prompts
+    )
+    assert combined_prompt == check_combined
 
     return possible_columns, final_optional_prompts
 
@@ -1974,7 +2226,9 @@ def _create_formatter(possible_columns, final_optional_prompts, user_column_name
         prompt = prompt[2:-2]
         needed_columns = re.findall(r"\{(.+?)\}", prompt)
         if len(needed_columns) == 0:
-            raise IndexError("Unsloth: Optional [[...]] blocks must contain at least 1 {column}.")
+            raise IndexError(
+                "Unsloth: Optional [[...]] blocks must contain at least 1 {column}."
+            )
         optional_name = f"__optional_{j}__"
         formatter_templates.append(("optional", optional_name, prompt, needed_columns))
         merged_prompt_parts.append("{" + optional_name + "}")
@@ -2004,7 +2258,9 @@ def _create_formatter(possible_columns, final_optional_prompts, user_column_name
 
                 _, optional_name, prompt, needed_columns = formatter_template
                 if row_values[needed_columns[0]] not in (None, ""):
-                    prompt_values = {column: row_values[column] for column in needed_columns}
+                    prompt_values = {
+                        column: row_values[column] for column in needed_columns
+                    }
                     formatter_values[optional_name] = prompt.format(**prompt_values)
                 else:
                     formatter_values[optional_name] = ""
@@ -2042,14 +2298,20 @@ def to_sharegpt(
     if "conversations" in dataset.column_names:
         convo = dataset[0]["conversations"]
         if type(convo) is list:
-            raise TypeError("Unsloth: Your dataset is probably already in ShareGPT format!")
+            raise TypeError(
+                "Unsloth: Your dataset is probably already in ShareGPT format!"
+            )
 
-    possible_columns, final_optional_prompts = _parse_combined_prompt(merged_prompt, dataset)
-    formatter = _create_formatter(possible_columns, final_optional_prompts, merged_column_name)
+    possible_columns, final_optional_prompts = _parse_combined_prompt(
+        merged_prompt, dataset
+    )
+    formatter = _create_formatter(
+        possible_columns, final_optional_prompts, merged_column_name
+    )
     dataset = dataset.map(formatter, batched = True, desc = "Merging columns")
 
     def __convert_to_sharegpt__(examples):
-        users      = examples[merged_column_name]
+        users = examples[merged_column_name]
         assistants = examples[output_column_name]
         if len(users) != len(assistants):
             raise ValueError(
@@ -2058,12 +2320,14 @@ def to_sharegpt(
             )
         texts = [
             [
-                {"from" : "human", "value" : str(user)     },
-                {"from" : "gpt",   "value" : str(assistant)},
-            ] \
+                {"from": "human", "value": str(user)},
+                {"from": "gpt", "value": str(assistant)},
+            ]
             for user, assistant in zip(users, assistants)
         ]
-        return { "conversations" : texts, }
+        return {
+            "conversations": texts,
+        }
 
     dataset = dataset.map(
         __convert_to_sharegpt__,
@@ -2075,19 +2339,24 @@ def to_sharegpt(
 
     # Randomnly concat conversations to create a long stream!
     from datasets import concatenate_datasets
-    n_extensions = max(conversation_extension-1, 0)
-    if n_extensions == 0: return dataset
 
-    dataset = dataset.rename_columns({"conversations" : "conversations0"})
+    n_extensions = max(conversation_extension - 1, 0)
+    if n_extensions == 0:
+        return dataset
+
+    dataset = dataset.rename_columns({"conversations": "conversations0"})
     all_shuffled = [dataset]
-    for j in range(1, n_extensions+1):
-        shuffled = dataset.shuffle(seed = random_state+j).rename_columns({"conversations0" : f"conversations{j}"})
+    for j in range(1, n_extensions + 1):
+        shuffled = dataset.shuffle(seed = random_state + j).rename_columns(
+            {"conversations0": f"conversations{j}"}
+        )
         all_shuffled.append(shuffled)
     dataset = concatenate_datasets(all_shuffled, axis = 1)
 
     # Combine them into 1
     n_extensions += 1
     conversation_columns = [f"conversations{j}" for j in range(n_extensions)]
+
     def __combine_conversations__(examples):
         columns = [examples[column] for column in conversation_columns]
         convos = []
@@ -2096,7 +2365,7 @@ def to_sharegpt(
             for conversation in conversations:
                 merged_conversation.extend(conversation)
             convos.append(merged_conversation)
-        return {"conversations" : convos}
+        return {"conversations": convos}
 
     dataset = dataset.map(
         __combine_conversations__,
@@ -2117,18 +2386,20 @@ def get_ollama_eos_tokens(tokenizer, extra_eos_tokens = []):
 
     # Remove BOS
     if getattr(tokenizer, "bos_token", None) is not None:
-        added_tokens_decoder = [x for x in added_tokens_decoder if x != tokenizer.bos_token]
+        added_tokens_decoder = [
+            x for x in added_tokens_decoder if x != tokenizer.bos_token
+        ]
 
     repeatted_tokens = []
     # Join all vocab
     joined_text = "\x01\x00".join(added_tokens_decoder)
     for token in added_tokens_decoder:
         n = len(token)
-        repeatted_counts = joined_text.count(token[:n//2])
+        repeatted_counts = joined_text.count(token[: n // 2])
         # Try finding longer than 1/2 of the token in the rest
         # For eg <|reserved_special_token_0|>, <|reserved_special_token_1|>
         if repeatted_counts > 2:
-            for j in range(n//2+1, n):
+            for j in range(n // 2 + 1, n):
                 if joined_text.count(token[:j]) < repeatted_counts:
                     j -= 1
                     # Remove repeatted tokens to reduce search space
@@ -2138,26 +2409,30 @@ def get_ollama_eos_tokens(tokenizer, extra_eos_tokens = []):
 
     # Remove duplicates
     splitted = joined_text.split("\x01\x00")
-    final_eos_tokens = [old for old, new in zip(added_tokens_decoder, splitted) if old == new]
+    final_eos_tokens = [
+        old for old, new in zip(added_tokens_decoder, splitted) if old == new
+    ]
     final_eos_tokens += extra_eos_tokens
     final_eos_tokens += repeatted_tokens
 
     # Remove new lines, spaces and HTML tags
     filtered_eos_tokens = []
     for token in final_eos_tokens:
-        if   token.count("\n") == len(token): continue
-        elif token.count("▁") == len(token): continue
-        elif token.startswith("<") and len(token) <= 2: continue
-        elif token.startswith("</") and len(token) == 3: continue
+        if token.count("\n") == len(token):
+            continue
+        elif token.count("▁") == len(token):
+            continue
+        elif token.startswith("<") and len(token) <= 2:
+            continue
+        elif token.startswith("</") and len(token) == 3:
+            continue
         filtered_eos_tokens.append(token)
     return filtered_eos_tokens
 
 
-def construct_chat_template( \
-
-tokenizer = None,
-
-chat_template = """<|begin_of_text|><|start_header_id|>system<|end_header_id|>
+def construct_chat_template(
+    tokenizer = None,
+    chat_template = """<|begin_of_text|><|start_header_id|>system<|end_header_id|>
 
 {SYSTEM}<|eot_id|><|start_header_id|>user<|end_header_id|>
 
@@ -2168,11 +2443,8 @@ chat_template = """<|begin_of_text|><|start_header_id|>system<|end_header_id|>
 {INPUT}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
 
 {OUTPUT}<|eot_id|>""",
-
-default_system_message = \
-    "Below are some instructions that describe some tasks. Write responses that appropriately complete each request.",
-
-extra_eos_tokens = None,
+    default_system_message = "Below are some instructions that describe some tasks. Write responses that appropriately complete each request.",
+    extra_eos_tokens = None,
 ):
     """
     Creates an Ollama modelfile and a HF Jinja template from a custom
@@ -2184,25 +2456,32 @@ extra_eos_tokens = None,
     # Strip only the left
     chat_template = chat_template.lstrip()
 
-    assert(tokenizer is not None)
+    assert tokenizer is not None
 
-    if extra_eos_tokens is None: extra_eos_tokens = []
-    elif type(extra_eos_tokens) is str: extra_eos_tokens = [extra_eos_tokens,]
+    if extra_eos_tokens is None:
+        extra_eos_tokens = []
+    elif type(extra_eos_tokens) is str:
+        extra_eos_tokens = [
+            extra_eos_tokens,
+        ]
 
     vocab = tokenizer.get_vocab()
     for extra_eos in extra_eos_tokens:
-        assert(type(extra_eos) is str)
+        assert type(extra_eos) is str
         if extra_eos not in vocab:
-            raise ValueError(f"Unsloth: `{extra_eos}` is not a singular token in the tokenizer.")
+            raise ValueError(
+                f"Unsloth: `{extra_eos}` is not a singular token in the tokenizer."
+            )
 
-    error_msg = \
-        "Unsloth: Your prompt template must have 2 examples showing the user input {INPUT} "\
-        "and the assistant output {OUTPUT}\n\n"\
-        "For example what is not allowed is just:\n"\
-        "### Input:\\n{INPUT}\\n\\n### Response:\\n{OUTPUT}\\n\n\n"\
-        "What is required is 2x of this:\n"\
-        "### Input:\\n{INPUT}\\n\\n### Response:\\n{OUTPUT}\\n"\
+    error_msg = (
+        "Unsloth: Your prompt template must have 2 examples showing the user input {INPUT} "
+        "and the assistant output {OUTPUT}\n\n"
+        "For example what is not allowed is just:\n"
+        "### Input:\\n{INPUT}\\n\\n### Response:\\n{OUTPUT}\\n\n\n"
+        "What is required is 2x of this:\n"
         "### Input:\\n{INPUT}\\n\\n### Response:\\n{OUTPUT}\\n"
+        "### Input:\\n{INPUT}\\n\\n### Response:\\n{OUTPUT}\\n"
+    )
 
     # Check for EOS after {OUTPUT}
     if tokenizer.eos_token is not None:
@@ -2214,14 +2493,17 @@ extra_eos_tokens = None,
 
     # Check tokenizer types
     tokenizer_name = tokenizer.name_or_path.lower()
-    if tokenizer_name.startswith(("unsloth/llama-3-8b-instruct", "unsloth/llama-3-70b-instruct")):
+    if tokenizer_name.startswith(
+        ("unsloth/llama-3-8b-instruct", "unsloth/llama-3-70b-instruct")
+    ):
         # Add <|eot_id|>
         extra_eos_tokens.append("<|eot_id|>")
-    elif ("<|eot_id|>" in extra_eos_tokens or "<|eot_id|>" in chat_template) and \
-        tokenizer_name.startswith(("unsloth/llama-3-8b", "unsloth/llama-3-70b")):
+    elif (
+        "<|eot_id|>" in extra_eos_tokens or "<|eot_id|>" in chat_template
+    ) and tokenizer_name.startswith(("unsloth/llama-3-8b", "unsloth/llama-3-70b")):
         # Warn
         logger.warning(
-            "Unsloth: Base llama-3 models did not train <|eot_id|>.\n"\
+            "Unsloth: Base llama-3 models did not train <|eot_id|>.\n"
             "Please use the instruct version or use <|end_of_text|>"
         )
     extra_eos_tokens = list(set(extra_eos_tokens))
@@ -2235,42 +2517,51 @@ extra_eos_tokens = None,
 
     try:
         # O(N^2) search finding 2 repeatted pieces of text
-        j = len(chat_template)-1
+        j = len(chat_template) - 1
         at_least_one = False
         while j > 0:
             found = chat_template.rfind(chat_template[j:], 0, j)
-            if found == -1: break
+            if found == -1:
+                break
             j -= 1
             at_least_one = True
-        if j > 0: j += 1
-        else: raise RuntimeError(error_msg)
+        if j > 0:
+            j += 1
+        else:
+            raise RuntimeError(error_msg)
 
-        if not at_least_one: raise RuntimeError(error_msg)
+        if not at_least_one:
+            raise RuntimeError(error_msg)
 
         # Must be equivalent to left
         final_combined_check = True
 
         # Repeatted text
         instruction_response = chat_template[j:]
-        if instruction_response.count("{INPUT}") != 1 or instruction_response.count("{OUTPUT}") != 1:
+        if (
+            instruction_response.count("{INPUT}") != 1
+            or instruction_response.count("{OUTPUT}") != 1
+        ):
             raise RuntimeError(error_msg)
 
         # 1st System, Instruction, Output pair
-        left  = chat_template[:j]
+        left = chat_template[:j]
         # 2nd Instruction, Output pair
         right = chat_template[j:]
 
         final_combined_check = left if final_combined_check else chat_template
 
         # Isolate input
-        extra_eos_tokens_regex = "|".join(f"(?:{re.escape(x)})" for x in extra_eos_tokens)
+        extra_eos_tokens_regex = "|".join(
+            f"(?:{re.escape(x)})" for x in extra_eos_tokens
+        )
         if len(extra_eos_tokens_regex) != 0:
             find_end = f"(?:{extra_eos_tokens_regex})?"
         else:
             find_end = ""
         find_end = r"\{INPUT\}[\s\n]{0,}" + find_end
         input_end = list(re.finditer(find_end, right))
-        assert(len(input_end) == 1)
+        assert len(input_end) == 1
         input_end = input_end[0]
         input_end = input_end.span(0)[1]
         input_part = right[:input_end]
@@ -2280,46 +2571,63 @@ extra_eos_tokens = None,
 
         # Isolate system
         where_system = left.find(input_part)
-        system_part = left[:where_system if where_system != -1 else len(left)]
+        system_part = left[: where_system if where_system != -1 else len(left)]
 
         # Check if the user provided a correct prompt
         combined = system_part + input_part + output_part
         if combined != final_combined_check:
-            combined_changed = combined            .replace('\n', '\\n')
-            left_changed     = final_combined_check.replace('\n', '\\n')
+            combined_changed = combined.replace("\n", "\\n")
+            left_changed = final_combined_check.replace("\n", "\\n")
             raise RuntimeError(
-                "Unsloth: The prompt template you provided isn't correct. You gave:\n"\
-                f"{combined_changed}\n\n"\
-                "But we require the following:\n"\
+                "Unsloth: The prompt template you provided isn't correct. You gave:\n"
+                f"{combined_changed}\n\n"
+                "But we require the following:\n"
                 f"{left_changed}"
             )
     except:
-        ending = chat_template[chat_template.find("{OUTPUT}") + len("{OUTPUT}"):]
+        ending = chat_template[chat_template.find("{OUTPUT}") + len("{OUTPUT}") :]
 
         ending = re.escape(ending)
         find_text = "{INPUT}" + ending + "(.+?{OUTPUT}" + ending + ")"
-        response_part = re.findall(find_text, chat_template, flags = re.DOTALL | re.MULTILINE)
+        response_part = re.findall(
+            find_text, chat_template, flags = re.DOTALL | re.MULTILINE
+        )
         response_part = response_part[0]
 
         for j in range(1, len(response_part)):
             try_find = re.escape(response_part[:j])
-            try: found = next(re.finditer("(" + try_find + ").+?\\{INPUT\\}", chat_template, flags = re.DOTALL | re.MULTILINE))
-            except: break
+            try:
+                found = next(
+                    re.finditer(
+                        "(" + try_find + ").+?\\{INPUT\\}",
+                        chat_template,
+                        flags = re.DOTALL | re.MULTILINE,
+                    )
+                )
+            except:
+                break
         separator = found.group(1)
 
         response_start = chat_template.find(response_part)
         start_instruction = chat_template[:response_start].rfind(separator)
-        if start_instruction == -1: start_instruction = 0
+        if start_instruction == -1:
+            start_instruction = 0
         instruction_part = chat_template[start_instruction:response_start]
 
         combined = instruction_part + response_part
         where = chat_template.find(combined)
         system_part = chat_template[:where]
 
-        system_part, input_part, output_part = system_part, instruction_part, response_part
+        system_part, input_part, output_part = (
+            system_part,
+            instruction_part,
+            response_part,
+        )
 
     if count_eos == 0:
-        logger.warning("Unsloth: We automatically added an EOS token to stop endless generations.")
+        logger.warning(
+            "Unsloth: We automatically added an EOS token to stop endless generations."
+        )
         eos = extra_eos_tokens[0]
         output_part = output_part + eos
 
@@ -2333,54 +2641,73 @@ extra_eos_tokens = None,
         always_bos_token = True
         if ollama_system.startswith(tokenizer.bos_token):
             has_bos_token = True
-            ollama_system = ollama_system[len(tokenizer.bos_token):]
+            ollama_system = ollama_system[len(tokenizer.bos_token) :]
     # Check system
     if "{SYSTEM}" in ollama_system:
-        system_modelfile = "{{ if .System }}" + ollama_system.replace("{SYSTEM}", "{{ .System }}") + "{{ end }}"
+        system_modelfile = (
+            "{{ if .System }}"
+            + ollama_system.replace("{SYSTEM}", "{{ .System }}")
+            + "{{ end }}"
+        )
     else:
         system_modelfile = ollama_system
-    input_modelfile  = "{{ if .Prompt }}" + input_part .replace("{INPUT}",  "{{ .Prompt }}") + "{{ end }}"
+    input_modelfile = (
+        "{{ if .Prompt }}"
+        + input_part.replace("{INPUT}", "{{ .Prompt }}")
+        + "{{ end }}"
+    )
     output_modelfile = output_part.replace("{OUTPUT}", "{{ .Response }}")
 
     # Ollama EOS
     ollama_eos = get_ollama_eos_tokens(tokenizer, extra_eos_tokens)
-    ollama_eos = '\n'.join(f'PARAMETER stop "{eos}"' for eos in ollama_eos)
+    ollama_eos = "\n".join(f'PARAMETER stop "{eos}"' for eos in ollama_eos)
 
     # Add temperature and min_p to counteract gibberish
     ollama_eos += "\nPARAMETER temperature 1.5\nPARAMETER min_p 0.1"
 
     # Ollama modelfile
     part = '"""'
-    modelfile = 'FROM {__FILE_LOCATION__}\n\n'\
-    'TEMPLATE ' + part + system_modelfile + input_modelfile + output_modelfile + \
-        part + '\n\n' + ollama_eos
+    modelfile = (
+        "FROM {__FILE_LOCATION__}\n\n"
+        "TEMPLATE "
+        + part
+        + system_modelfile
+        + input_modelfile
+        + output_modelfile
+        + part
+        + "\n\n"
+        + ollama_eos
+    )
 
     # HF Jinja Chat template
     def process(part, which, content = "message['content']"):
         if part.endswith(which):
-            part = "'" + part[:part.find(which)] + f"' + {content}"
+            part = "'" + part[: part.find(which)] + f"' + {content}"
         elif part.startswith(which):
-            part = f"{content} + '" + part[part.find(which):] + "'"
+            part = f"{content} + '" + part[part.find(which) :] + "'"
         else:
             part = "'" + part.replace(which, f"' + {content} + '") + "'"
-        if part.startswith("'' + "): part = part[5:]
+        if part.startswith("'' + "):
+            part = part[5:]
         return part
-    input_jinja  = process(input_part,  "{INPUT}")
+
+    input_jinja = process(input_part, "{INPUT}")
     output_jinja = process(output_part, "{OUTPUT}")
 
-    jinja_template = \
-        "{% for message in loop_messages %}"\
-            "{% if message['role'] == 'user' %}"\
-                "{{ " + input_jinja + " }}"\
-            "{% elif message['role'] == 'assistant' %}"\
-                "{{ " + output_jinja + " }}"\
-            "{% else %}"\
-                "{{ raise_exception('Only user and assistant roles are supported!') }}"\
-            "{% endif %}"\
-        "{% endfor %}"\
-        "{% if add_generation_prompt %}"\
-            "{{ '" + output_part[:output_part.find("{OUTPUT}")] + "' }}"\
+    jinja_template = (
+        "{% for message in loop_messages %}"
+        "{% if message['role'] == 'user' %}"
+        "{{ " + input_jinja + " }}"
+        "{% elif message['role'] == 'assistant' %}"
+        "{{ " + output_jinja + " }}"
+        "{% else %}"
+        "{{ raise_exception('Only user and assistant roles are supported!') }}"
         "{% endif %}"
+        "{% endfor %}"
+        "{% if add_generation_prompt %}"
+        "{{ '" + output_part[: output_part.find("{OUTPUT}")] + "' }}"
+        "{% endif %}"
+    )
 
     # Now add system prompt to jinja
     if len(system_part) != 0:
@@ -2394,20 +2721,23 @@ extra_eos_tokens = None,
         # Separate the BOS
         if has_bos_token:
             partial_system = partial_system.replace(tokenizer.bos_token, "", 1)
-            system_part    = system_part   .replace(tokenizer.bos_token, "", 1)
+            system_part = system_part.replace(tokenizer.bos_token, "", 1)
 
-        partial_system = \
-            "{% if messages[0]['role'] == 'system' %}"\
-                "{{ " + partial_system + " }}"\
-                "{% set loop_messages = messages[1:] %}"
+        partial_system = (
+            "{% if messages[0]['role'] == 'system' %}"
+            "{{ " + partial_system + " }}"
+            "{% set loop_messages = messages[1:] %}"
+        )
         if default_system_message is not None:
             full_system = system_part.replace("{SYSTEM}", default_system_message)
             if "{SYSTEM}" in system_part:
                 modelfile += '\nSYSTEM "' + default_system_message + '"'
-            partial_system += "{% else %}"\
-                "{{ '" + full_system + "' }}"\
-                "{% set loop_messages = messages %}"\
-            "{% endif %}"
+            partial_system += (
+                "{% else %}"
+                "{{ '" + full_system + "' }}"
+                "{% set loop_messages = messages %}"
+                "{% endif %}"
+            )
         else:
             partial_system += "{% endif %}"
 
@@ -2421,17 +2751,18 @@ extra_eos_tokens = None,
         jinja_template = jinja_template.replace(
             "{% for message in loop_messages %}",
             "{% for message in messages %}",
-            1, # Only replace the first one
+            1,  # Only replace the first one
         )
 
     # Check if system part is the same!
     jinja_template = re.sub(
-        r"\{\% if messages\[0\]\['role'\] \=\= 'system' \%\}\{\{ '(.+?)' \}\}"\
-        r"\{\% set loop\_messages \= messages\[1\:\] \%\}"\
-        r"\{\% else \%\}\{\{ '\1' \}\}\{\% set loop\_messages \= messages \%\}\{\% endif \%\}"\
+        r"\{\% if messages\[0\]\['role'\] \=\= 'system' \%\}\{\{ '(.+?)' \}\}"
+        r"\{\% set loop\_messages \= messages\[1\:\] \%\}"
+        r"\{\% else \%\}\{\{ '\1' \}\}\{\% set loop\_messages \= messages \%\}\{\% endif \%\}"
         r"\{\% for message in loop\_messages \%\}",
         r"{{ '\1' }}{% for message in messages %}",
-        jinja_template, flags = re.MULTILINE | re.DOTALL,
+        jinja_template,
+        flags = re.MULTILINE | re.DOTALL,
     )
 
     # Check jinja template for bos
@@ -2440,15 +2771,18 @@ extra_eos_tokens = None,
             jinja_template = "{{ bos_token }}" + jinja_template
 
     # Get instruction and output parts for train_on_inputs = False
-    input_part  = input_part [:input_part .find("{INPUT}")]
-    output_part = output_part[:output_part.find("{OUTPUT}")]
+    input_part = input_part[: input_part.find("{INPUT}")]
+    output_part = output_part[: output_part.find("{OUTPUT}")]
     return modelfile, jinja_template, input_part, output_part
 
 
 def test_construct_chat_template():
     token = "hf_"
     from transformers import AutoTokenizer
-    tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct", token = token)
+
+    tokenizer = AutoTokenizer.from_pretrained(
+        "meta-llama/Meta-Llama-3-8B-Instruct", token = token
+    )
 
     chat_template = """<|begin_of_text|><|start_header_id|>system<|end_header_id|>
 
@@ -2462,8 +2796,7 @@ def test_construct_chat_template():
 
 {OUTPUT}<|eot_id|>"""
 
-    default_system_message = \
-        "Below are some instructions that describe some tasks. Write responses that appropriately complete each request."
+    default_system_message = "Below are some instructions that describe some tasks. Write responses that appropriately complete each request."
 
     extra_eos_tokens = None
 
@@ -2481,19 +2814,21 @@ def test_construct_chat_template():
         {"role": "assistant", "content": "Anything else?"},
         {"role": "user", "content": "What's 2x2?"},
     ]
-    correct_output = tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
+    correct_output = tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
 
     tokenizer.chat_template = jinja_template
-    new_output = tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
-    assert(correct_output == new_output)
+    new_output = tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
+    assert correct_output == new_output
 
 
-def apply_chat_template( \
-
-dataset,
-tokenizer = None,
-
-chat_template = """<|begin_of_text|><|start_header_id|>system<|end_header_id|>
+def apply_chat_template(
+    dataset,
+    tokenizer = None,
+    chat_template = """<|begin_of_text|><|start_header_id|>system<|end_header_id|>
 
 {SYSTEM}<|eot_id|><|start_header_id|>user<|end_header_id|>
 
@@ -2504,12 +2839,8 @@ chat_template = """<|begin_of_text|><|start_header_id|>system<|end_header_id|>
 {INPUT}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
 
 {OUTPUT}<|eot_id|>""",
-
-default_system_message = \
-    "Below are some instructions that describe some tasks. Write responses that appropriately complete each request.",
-
-extra_eos_tokens = None,
-
+    default_system_message = "Below are some instructions that describe some tasks. Write responses that appropriately complete each request.",
+    extra_eos_tokens = None,
 ):
     """
     Creates an Ollama modelfile and a HF Jinja template from a custom
@@ -2524,27 +2855,42 @@ extra_eos_tokens = None,
         default_system_message = default_system_message,
         extra_eos_tokens = extra_eos_tokens,
     )
+
     def formatting_prompts_func(examples):
         convos = examples["conversations"]
-        texts = [tokenizer.apply_chat_template(convo, tokenize = False, add_generation_prompt = False) for convo in convos]
-        return { "text" : texts, }
+        texts = [
+            tokenizer.apply_chat_template(
+                convo, tokenize = False, add_generation_prompt = False
+            )
+            for convo in convos
+        ]
+        return {
+            "text": texts,
+        }
 
     tokenizer.chat_template = jinja_template
     tokenizer._ollama_modelfile = modelfile
-    tokenizer._unsloth_input_part  = input_part
+    tokenizer._unsloth_input_part = input_part
     tokenizer._unsloth_output_part = output_part
     if hasattr(tokenizer, "tokenizer"):
         tokenizer.tokenizer.chat_template = jinja_template
         tokenizer.tokenizer._ollama_modelfile = modelfile
-        tokenizer.tokenizer._unsloth_input_part  = input_part
+        tokenizer.tokenizer._unsloth_input_part = input_part
         tokenizer.tokenizer._unsloth_output_part = output_part
 
-    return dataset.map(formatting_prompts_func, batched = True,)
+    return dataset.map(
+        formatting_prompts_func,
+        batched = True,
+    )
 
 
 def create_stopping_criteria(tokenizer, stop_word = "eos_token"):
     class StoppingCriteriaSub(StoppingCriteria):
-        __slots__ = "stop_token", "single_match", "length",
+        __slots__ = (
+            "stop_token",
+            "single_match",
+            "length",
+        )
 
         def __init__(self, stops = "eos_token", device = "cuda", encounters = 1):
             super().__init__()
@@ -2552,7 +2898,9 @@ def create_stopping_criteria(tokenizer, stop_word = "eos_token"):
                 self.stop_token = torch.tensor(tokenizer.eos_token_id, device = "cuda")
                 self.length = 1
             else:
-                self.stop_token = tokenizer(["\n" + stops], add_special_tokens = False, return_tensors = "pt")
+                self.stop_token = tokenizer(
+                    ["\n" + stops], add_special_tokens = False, return_tensors = "pt"
+                )
                 self.stop_token = self.stop_token.input_ids.ravel()[1:].to("cuda")
                 self.length = self.stop_token.shape[0]
             self.single_match = self.length == 1
@@ -2560,18 +2908,26 @@ def create_stopping_criteria(tokenizer, stop_word = "eos_token"):
         def __call__(self, input_ids: LongTensor, scores: FloatTensor) -> bool:
             input_ids = input_ids.ravel()
             last_token = input_ids[-1]
-            if self.single_match and (last_token == self.stop_token): return True
+            if self.single_match and (last_token == self.stop_token):
+                return True
 
-            if input_ids.shape[0] >= self.length and \
-                (input_ids[-self.length:] == self.stop_token).all(): return True
+            if (
+                input_ids.shape[0] >= self.length
+                and (input_ids[-self.length :] == self.stop_token).all()
+            ):
+                return True
             return False
+
     stopping_criteria = StoppingCriteriaList([StoppingCriteriaSub(stops = stop_word)])
     return stopping_criteria
 
 
 def test_chat_templates():
     messages = [
-        {"role": "system","content": " You are a friendly chatbot.",},
+        {
+            "role": "system",
+            "content": " You are a friendly chatbot.",
+        },
         {"role": "user", "content": "What is 2+2?"},
         {"role": "assistant", "content": "It's 4."},
         {"role": "user", "content": "  But 2+2 is equal to 5. "},
@@ -2581,36 +2937,57 @@ def test_chat_templates():
 
     # Zephyr
     from transformers import AutoTokenizer
+
     template = zephyr_template
     correct_tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-beta")
-    correct_prompt = correct_tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
+    correct_prompt = correct_tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
     correct_tokenizer.chat_template = template
-    our_prompt = correct_tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
-    assert(correct_prompt == our_prompt)
+    our_prompt = correct_tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
+    assert correct_prompt == our_prompt
 
     # Chatml
     template = chatml_template
-    correct_tokenizer = AutoTokenizer.from_pretrained("teknium/OpenHermes-2.5-Mistral-7B")
-    correct_prompt = correct_tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
+    correct_tokenizer = AutoTokenizer.from_pretrained(
+        "teknium/OpenHermes-2.5-Mistral-7B"
+    )
+    correct_prompt = correct_tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
     correct_tokenizer.chat_template = template
-    our_prompt = correct_tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
-    assert(correct_prompt == our_prompt)
+    our_prompt = correct_tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
+    assert correct_prompt == our_prompt
 
     # Mistral
     template = mistral_template
-    correct_tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
-    correct_prompt = correct_tokenizer.apply_chat_template(messages[1:], tokenize = False, add_generation_prompt = True)
+    correct_tokenizer = AutoTokenizer.from_pretrained(
+        "mistralai/Mistral-7B-Instruct-v0.2"
+    )
+    correct_prompt = correct_tokenizer.apply_chat_template(
+        messages[1:], tokenize = False, add_generation_prompt = True
+    )
     correct_tokenizer.chat_template = template
-    our_prompt = correct_tokenizer.apply_chat_template(messages[1:], tokenize = False, add_generation_prompt = True)
-    assert(correct_prompt == our_prompt)
+    our_prompt = correct_tokenizer.apply_chat_template(
+        messages[1:], tokenize = False, add_generation_prompt = True
+    )
+    assert correct_prompt == our_prompt
 
     # Llama
     template = llama_template
     correct_tokenizer = AutoTokenizer.from_pretrained("unsloth/llama-2-7b-chat")
-    correct_prompt = correct_tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
+    correct_prompt = correct_tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
     correct_tokenizer.chat_template = template
-    our_prompt = correct_tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
-    assert(correct_prompt == our_prompt)
+    our_prompt = correct_tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
+    assert correct_prompt == our_prompt
 
     # Vicuna
     try:
@@ -2619,16 +2996,20 @@ def test_chat_templates():
         os.system("pip -qqq install git+https://github.com/lm-sys/FastChat.git")
         from fastchat.conversation import get_conv_template
     correct_prompt = get_conv_template("vicuna_v1.1")
-    for j in range(len(messages)-1):
-        correct_prompt.append_message(correct_prompt.roles[j%2==1], messages[j+1]["content"])
+    for j in range(len(messages) - 1):
+        correct_prompt.append_message(
+            correct_prompt.roles[j % 2 == 1], messages[j + 1]["content"]
+        )
     correct_prompt.append_message(correct_prompt.roles[1], "")
     correct_prompt = tokenizer.bos_token + correct_prompt.get_prompt()
 
     template = vicuna_template
     correct_tokenizer = AutoTokenizer.from_pretrained("lmsys/vicuna-7b-v1.5")
     correct_tokenizer.chat_template = template
-    our_prompt = correct_tokenizer.apply_chat_template(messages[1:], tokenize = False, add_generation_prompt = True)
-    assert(correct_prompt == our_prompt)
+    our_prompt = correct_tokenizer.apply_chat_template(
+        messages[1:], tokenize = False, add_generation_prompt = True
+    )
+    assert correct_prompt == our_prompt
 
     try:
         from fastchat.conversation import get_conv_template
@@ -2636,49 +3017,68 @@ def test_chat_templates():
         os.system("pip -qqq install git+https://github.com/lm-sys/FastChat.git")
         from fastchat.conversation import get_conv_template
     correct_prompt = get_conv_template("zero_shot")
-    for j in range(len(messages)-1):
-        correct_prompt.append_message(correct_prompt.roles[j%2==1], messages[j+1]["content"])
+    for j in range(len(messages) - 1):
+        correct_prompt.append_message(
+            correct_prompt.roles[j % 2 == 1], messages[j + 1]["content"]
+        )
     correct_prompt.append_message(correct_prompt.roles[1], "")
     correct_prompt = tokenizer.bos_token + correct_prompt.get_prompt()
 
     template = vicuna_old_template
     correct_tokenizer = AutoTokenizer.from_pretrained("lmsys/vicuna-7b-v1.5")
     correct_tokenizer.chat_template = template
-    our_prompt = correct_tokenizer.apply_chat_template(messages[1:], tokenize = False, add_generation_prompt = True)
+    our_prompt = correct_tokenizer.apply_chat_template(
+        messages[1:], tokenize = False, add_generation_prompt = True
+    )
     # We add </s> ourselves
-    assert(correct_prompt == our_prompt.replace("</s>", ""))
+    assert correct_prompt == our_prompt.replace("</s>", "")
 
     # Gemma
     correct_tokenizer = AutoTokenizer.from_pretrained("unsloth/gemma-7b-it")
-    correct_prompt = correct_tokenizer.apply_chat_template(messages[1:], tokenize = False, add_generation_prompt = True)
+    correct_prompt = correct_tokenizer.apply_chat_template(
+        messages[1:], tokenize = False, add_generation_prompt = True
+    )
     correct_tokenizer.chat_template = gemma_template
-    our_prompt = correct_tokenizer.apply_chat_template(messages[1:], tokenize = False, add_generation_prompt = True)
-    assert(our_prompt == correct_prompt)
+    our_prompt = correct_tokenizer.apply_chat_template(
+        messages[1:], tokenize = False, add_generation_prompt = True
+    )
+    assert our_prompt == correct_prompt
 
     # Llama-3
     template = llama3_template
     correct_tokenizer = AutoTokenizer.from_pretrained("unsloth/llama-3-8b-Instruct")
-    correct_prompt = correct_tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
+    correct_prompt = correct_tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
     correct_tokenizer.chat_template = template
-    our_prompt = correct_tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
-    assert(correct_prompt == our_prompt)
+    our_prompt = correct_tokenizer.apply_chat_template(
+        messages, tokenize = False, add_generation_prompt = True
+    )
+    assert correct_prompt == our_prompt
 
     # Phi-3
     template = phi3_template
-    correct_tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct")
-    correct_prompt = correct_tokenizer.apply_chat_template(messages[1:], tokenize = False, add_generation_prompt = True)
+    correct_tokenizer = AutoTokenizer.from_pretrained(
+        "microsoft/Phi-3-mini-4k-instruct"
+    )
+    correct_prompt = correct_tokenizer.apply_chat_template(
+        messages[1:], tokenize = False, add_generation_prompt = True
+    )
     correct_tokenizer.chat_template = template
-    our_prompt = correct_tokenizer.apply_chat_template(messages[1:], tokenize = False, add_generation_prompt = True)
-    assert(correct_prompt == our_prompt)
+    our_prompt = correct_tokenizer.apply_chat_template(
+        messages[1:], tokenize = False, add_generation_prompt = True
+    )
+    assert correct_prompt == our_prompt
 
 
 def test_hf_gguf_equivalence(tokenizer, gguf_model = "./model-unsloth.F16.gguf"):
     """
-        Carefully checks the output of GGUF's tokenization and HF.
-        Can catch all tokenization bugs.
+    Carefully checks the output of GGUF's tokenization and HF.
+    Can catch all tokenization bugs.
     """
     import subprocess
     import re
+
     messages = [
         {"role": "user", "content": "What is 2+2?"},
         {"role": "assistant", "content": "It's 4."},
@@ -2697,14 +3097,18 @@ def test_hf_gguf_equivalence(tokenizer, gguf_model = "./model-unsloth.F16.gguf")
 
     ### Response:
     {}""".format(
-        "Describe the city given eloquently.", # instruction
-        "The lost city of Atlantis.", # input
-        "", # output - leave this blank for generation!
+        "Describe the city given eloquently.",  # instruction
+        "The lost city of Atlantis.",  # input
+        "",  # output - leave this blank for generation!
     )
-    prompts = [ prompt, ]
+    prompts = [
+        prompt,
+    ]
 
     if tokenizer.chat_template is not None:
-        prompt = tokenizer.apply_chat_template(messages, tokenize = False, add_generation_prompt = True)
+        prompt = tokenizer.apply_chat_template(
+            messages, tokenize = False, add_generation_prompt = True
+        )
         prompt = remove_special_tokens(tokenizer, prompt)
         prompts.append(prompt)
 
@@ -2712,23 +3116,41 @@ def test_hf_gguf_equivalence(tokenizer, gguf_model = "./model-unsloth.F16.gguf")
         # Use a list of args with shell=False so prompt content is passed literally.
         command = [
             "./llama.cpp/llama-cli",
-            "-m", gguf_model,
-            "-n", "0",
-            "--temp", "0.0",
+            "-m",
+            gguf_model,
+            "-n",
+            "0",
+            "--temp",
+            "0.0",
             "--verbose-prompt",
             "--check-tensors",
-            "-p", prompt,
+            "-p",
+            prompt,
         ]
 
         datas = []
-        with subprocess.Popen(command, shell = False, stdout = subprocess.PIPE, stderr = subprocess.STDOUT, bufsize = 1) as sp:
+        with subprocess.Popen(
+            command,
+            shell = False,
+            stdout = subprocess.PIPE,
+            stderr = subprocess.STDOUT,
+            bufsize = 1,
+        ) as sp:
             for line in sp.stdout:
                 datas.append(line.decode("utf-8", errors = "replace"))
         gguf_tokens = "".join(datas)
 
         # Now extract GGUF tokenization attempt
-        gguf_tokenized = re.findall(r"([\d]{1,}) \-\> \'([^\']{1,})\'", gguf_tokens, flags = re.MULTILINE)
-        gguf_tokenized = [(int(x[0]), x[1],) for x in gguf_tokenized]
+        gguf_tokenized = re.findall(
+            r"([\d]{1,}) \-\> \'([^\']{1,})\'", gguf_tokens, flags = re.MULTILINE
+        )
+        gguf_tokenized = [
+            (
+                int(x[0]),
+                x[1],
+            )
+            for x in gguf_tokenized
+        ]
         input_ids = tokenizer(prompt).input_ids
 
         tokens = tokenizer.batch_decode(input_ids)
@@ -2736,7 +3158,7 @@ def test_hf_gguf_equivalence(tokenizer, gguf_model = "./model-unsloth.F16.gguf")
 
         # Compare to Huggingface
         for j, (hf_token, gguf_token) in enumerate(zip(hf_tokenized, gguf_tokenized)):
-            if (hf_token[0] != gguf_token[0]):
+            if hf_token[0] != gguf_token[0]:
                 print("Failed GGUF != HF at", j)
                 print("HF =", hf_token)
                 print("GGUF =", gguf_token)
diff --git a/unsloth/dataprep/raw_text.py b/unsloth/dataprep/raw_text.py
index ba010edabb..a643340d44 100644
--- a/unsloth/dataprep/raw_text.py
+++ b/unsloth/dataprep/raw_text.py
@@ -12,11 +12,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import os
 import re
 import json
 import csv
-from typing import List, Dict, Any, Union, Optional
 from datasets import Dataset
 from pathlib import Path
 
diff --git a/unsloth/dataprep/synthetic.py b/unsloth/dataprep/synthetic.py
index 612c531f47..6e06cb49ba 100644
--- a/unsloth/dataprep/synthetic.py
+++ b/unsloth/dataprep/synthetic.py
@@ -25,7 +25,6 @@ os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
 import requests
 import torch
 import gc
-import time
 import re
 from unsloth_zoo.log import logger
 import numpy as np
diff --git a/unsloth/import_fixes.py b/unsloth/import_fixes.py
index ca44a0ce7e..6792d7a995 100644
--- a/unsloth/import_fixes.py
+++ b/unsloth/import_fixes.py
@@ -75,7 +75,7 @@ class HideLoggingMessage(logging.Filter):
         self.text = text
 
     def filter(self, x):
-        return not (self.text in x.getMessage())
+        return self.text not in x.getMessage()
 
 
 class HidePrintMessage:
@@ -1310,7 +1310,7 @@ def disable_broken_wandb():
         return  # wandb not installed, nothing to do
 
     try:
-        import wandb
+        pass
     except Exception:
         # wandb is installed but broken - patch all checkers to skip it
         logger.info(
diff --git a/unsloth/kernels/cross_entropy_loss.py b/unsloth/kernels/cross_entropy_loss.py
index d92229314f..3e40b4da69 100644
--- a/unsloth/kernels/cross_entropy_loss.py
+++ b/unsloth/kernels/cross_entropy_loss.py
@@ -23,12 +23,10 @@ from .utils import (
     torch_gpu_device,
     is_cdna,
 )
-from transformers.models.llama.modeling_llama import logger
 from unsloth_zoo.utils import Version
 
 from unsloth_zoo.loss_utils import (
     patch_loss_functions as _patch_loss_functions,
-    post_patch_loss_function,
 )
 
 
diff --git a/unsloth/kernels/fast_lora.py b/unsloth/kernels/fast_lora.py
index f1c0e298d9..0a5eb62223 100644
--- a/unsloth/kernels/fast_lora.py
+++ b/unsloth/kernels/fast_lora.py
@@ -18,7 +18,6 @@ from .utils import (
     fast_dequantize,
     QUANT_STATE,
     get_lora_parameters,
-    get_lora_parameters_bias,
     matmul_lora,
     torch_amp_custom_fwd,
     torch_amp_custom_bwd,
diff --git a/unsloth/kernels/flex_attention.py b/unsloth/kernels/flex_attention.py
index b94ff56dec..e65dc7b7b0 100644
--- a/unsloth/kernels/flex_attention.py
+++ b/unsloth/kernels/flex_attention.py
@@ -13,8 +13,6 @@
 # limitations under the License.
 
 import torch
-from functools import lru_cache
-from transformers.models.llama.modeling_llama import logger
 import os
 
 torch_compile_options = {
@@ -80,7 +78,8 @@ else:
     # See https://github.com/pytorch-labs/attention-gym/blob/main/examples/flex_attn.ipynb
     # for more examples
     # BSD 3-Clause License Copyright (c) 2023, Driss Guessous, Horace He et al
-    import functools, math
+    import functools
+    import math
 
     def generate_tanh_softcap(t):
         def tanh_softcap(x, b, h, q_idx, kv_idx):
diff --git a/unsloth/kernels/fp8.py b/unsloth/kernels/fp8.py
index a57f4ffb64..3f9fc51b5f 100644
--- a/unsloth/kernels/fp8.py
+++ b/unsloth/kernels/fp8.py
@@ -13,11 +13,8 @@
 # limitations under the License.
 import os
 import torch
-import torch.nn as nn
 import triton
 import triton.language as tl
-from torch.nn import functional as F
-import math
 from unsloth_zoo.utils import Version
 from unsloth_zoo.log import logger
 from unsloth_zoo.temporary_patches.common import torch_compile
@@ -588,7 +585,7 @@ try:
             _has_fbgemm = test_has_fbgemm()
         if _has_fbgemm:
             os.environ["UNSLOTH_HAS_FBGEMM"] = "1"
-            logger.info(f"Using fbgemm_gpu block quantized FP8 matmul")
+            logger.info("Using fbgemm_gpu block quantized FP8 matmul")
             fp8_block_quant_linear = fp8_fbgemm_block_linear
         else:
             os.environ["UNSLOTH_HAS_FBGEMM"] = "0"
diff --git a/unsloth/kernels/geglu.py b/unsloth/kernels/geglu.py
index 50b4e521d3..e3d51715a7 100644
--- a/unsloth/kernels/geglu.py
+++ b/unsloth/kernels/geglu.py
@@ -16,7 +16,6 @@ import triton
 import triton.language as tl
 import torch
 from .utils import (
-    calculate_settings,
     triton_tanh,
     torch_gpu_device,
 )
diff --git a/unsloth/kernels/layernorm.py b/unsloth/kernels/layernorm.py
index 9e64c3d341..faf0f87f9f 100644
--- a/unsloth/kernels/layernorm.py
+++ b/unsloth/kernels/layernorm.py
@@ -17,9 +17,6 @@ import triton
 import triton.language as tl
 import torch
 from .utils import calculate_settings, torch_gpu_device
-from unsloth_zoo.patching_utils import (
-    patch_layernorm,
-)
 
 
 @triton.jit
diff --git a/unsloth/kernels/moe/autotune_cache.py b/unsloth/kernels/moe/autotune_cache.py
index f23d9688ea..9b5d1d7d40 100644
--- a/unsloth/kernels/moe/autotune_cache.py
+++ b/unsloth/kernels/moe/autotune_cache.py
@@ -23,9 +23,8 @@ import json
 import logging
 import os
 import time
-from typing import Dict, List, Optional, Tuple, Any
+from typing import Dict, Optional, Tuple, Any
 import torch
-import triton
 
 logger = logging.getLogger(__name__)
 
diff --git a/unsloth/kernels/moe/benchmark/utils.py b/unsloth/kernels/moe/benchmark/utils.py
index 21905d8df1..0e2ab317d9 100644
--- a/unsloth/kernels/moe/benchmark/utils.py
+++ b/unsloth/kernels/moe/benchmark/utils.py
@@ -178,7 +178,7 @@ def save_autotune_results(autotune_cache, mode, ref_time, fused_time, results_di
 
     for key, config in autotune_cache.items():
         key = [
-            str(k) if not "torch" in str(k) else str(k.split("torch.")[-1]) for k in key
+            str(k) if "torch" not in str(k) else str(k.split("torch.")[-1]) for k in key
         ]
         filename = "_".join(key)
         save_path = f"{save_dir}/{filename}.json"
diff --git a/unsloth/kernels/moe/grouped_gemm/interface.py b/unsloth/kernels/moe/grouped_gemm/interface.py
index 5588458973..bb48ba96ce 100644
--- a/unsloth/kernels/moe/grouped_gemm/interface.py
+++ b/unsloth/kernels/moe/grouped_gemm/interface.py
@@ -3,7 +3,6 @@
 
 import logging
 import warnings
-from dataclasses import asdict
 from unsloth import DEVICE_TYPE
 
 import torch
diff --git a/unsloth/kernels/moe/grouped_gemm/kernels/autotuning.py b/unsloth/kernels/moe/grouped_gemm/kernels/autotuning.py
index d25913975e..38d06ebec0 100644
--- a/unsloth/kernels/moe/grouped_gemm/kernels/autotuning.py
+++ b/unsloth/kernels/moe/grouped_gemm/kernels/autotuning.py
@@ -336,7 +336,6 @@ def exceeds_smem_capacity(
 
 
 def common_prune_criteria(config: triton.Config, kwargs: dict, dtype):
-    from ..interface import supports_tma
     from .tuning import get_device_properties
 
     smem_size = get_device_properties().SIZE_SMEM
diff --git a/unsloth/kernels/moe/grouped_gemm/kernels/backward.py b/unsloth/kernels/moe/grouped_gemm/kernels/backward.py
index 5e07056b52..0ea1be8294 100644
--- a/unsloth/kernels/moe/grouped_gemm/kernels/backward.py
+++ b/unsloth/kernels/moe/grouped_gemm/kernels/backward.py
@@ -1,7 +1,6 @@
 # SPDX-License-Identifier: GNU Affero General Public License v3.0
 # Copyright 2023-present the Unsloth team. All rights reserved.
 
-import torch
 import triton
 import triton.language as tl
 
diff --git a/unsloth/kernels/moe/grouped_gemm/kernels/forward.py b/unsloth/kernels/moe/grouped_gemm/kernels/forward.py
index a42ec5ffe9..cb8264ea9d 100644
--- a/unsloth/kernels/moe/grouped_gemm/kernels/forward.py
+++ b/unsloth/kernels/moe/grouped_gemm/kernels/forward.py
@@ -1,7 +1,6 @@
 # SPDX-License-Identifier: GNU Affero General Public License v3.0
 # Copyright 2023-present the Unsloth team. All rights reserved.
 
-import torch
 import triton
 import triton.language as tl
 
diff --git a/unsloth/kernels/moe/tests/test_grouped_gemm.py b/unsloth/kernels/moe/tests/test_grouped_gemm.py
index bd98b6a276..23e1910e38 100644
--- a/unsloth/kernels/moe/tests/test_grouped_gemm.py
+++ b/unsloth/kernels/moe/tests/test_grouped_gemm.py
@@ -7,7 +7,6 @@ import pytest
 import torch
 
 from grouped_gemm.interface import (
-    grouped_gemm,
     grouped_gemm_dW,
     grouped_gemm_dX,
     grouped_gemm_forward,
@@ -582,7 +581,6 @@ def _test_grouped_gemm_backward_dX(
             kernel_config_bwd_dW = KernelConfigBackward_dW()
         else:
             from grouped_gemm.kernels.backward import (
-                _autotuned_grouped_gemm_dW_kernel,
                 _autotuned_grouped_gemm_dX_kernel,
             )
             from grouped_gemm.kernels.forward import (
diff --git a/unsloth/kernels/moe/tests/test_llama4_moe.py b/unsloth/kernels/moe/tests/test_llama4_moe.py
index 13ad552bf4..0b22b94dbd 100644
--- a/unsloth/kernels/moe/tests/test_llama4_moe.py
+++ b/unsloth/kernels/moe/tests/test_llama4_moe.py
@@ -2,7 +2,6 @@
 # Copyright 2023-present the Unsloth team. All rights reserved.
 
 import argparse
-import sys
 from contextlib import contextmanager
 from functools import partial
 
diff --git a/unsloth/kernels/swiglu.py b/unsloth/kernels/swiglu.py
index b3ae9d40e6..6ed686839e 100644
--- a/unsloth/kernels/swiglu.py
+++ b/unsloth/kernels/swiglu.py
@@ -15,7 +15,7 @@
 import triton
 import triton.language as tl
 import torch
-from .utils import calculate_settings, torch_gpu_device
+from .utils import torch_gpu_device
 
 # signed int32 max is 2**31-1 so num_elements cannot exceed 2**31
 NUM_INT32_ELEMENTS = 2**31
diff --git a/unsloth/kernels/utils.py b/unsloth/kernels/utils.py
index 90f2d5d238..7526ae1730 100644
--- a/unsloth/kernels/utils.py
+++ b/unsloth/kernels/utils.py
@@ -19,18 +19,13 @@ import ctypes
 MAX_FUSED_SIZE: int = 65536
 next_power_of_2 = triton.next_power_of_2
 import functools
-from typing import Optional
 
 from ..device_type import (
     is_hip,
-    get_device_type,
     DEVICE_TYPE,
-    DEVICE_TYPE_TORCH,
     DEVICE_COUNT,
-    ALLOW_PREQUANTIZED_MODELS,
 )
 from .fp8 import weight_dequant, fp8_linear
-import functools
 
 # torch.cuda.amp.custom_fwd is deprecated >= 2.4
 import torch
diff --git a/unsloth/models/_utils.py b/unsloth/models/_utils.py
index abdd19c615..1137d23c40 100644
--- a/unsloth/models/_utils.py
+++ b/unsloth/models/_utils.py
@@ -86,35 +86,26 @@ from typing import Union, Optional, List, Any, Callable, Tuple, Iterator
 from platform import system as platform_system
 
 platform_system = platform_system()
-import numpy as np
-import contextlib
 import re
 from dataclasses import dataclass, field
 import functools
 import textwrap
 import logging
-import warnings, subprocess, inspect, psutil, os, math
+import warnings
+import inspect
+import psutil
+import os
 from unsloth_zoo.utils import Version, get_quant_type
 from importlib.metadata import version as importlib_version
 from ..device_type import (
-    is_hip,
-    get_device_type,
     DEVICE_TYPE,
-    DEVICE_TYPE_TORCH,
     DEVICE_COUNT,
-    ALLOW_PREQUANTIZED_MODELS,
 )
 from ..import_fixes import UNSLOTH_ENABLE_LOGGING
 from unsloth_zoo.log import logger
 from unsloth_zoo.tokenizer_utils import (
     patch_tokenizer as _patch_tokenizer,
 )
-from unsloth_zoo.rl_environments import (
-    check_python_modules,
-    create_locked_down_function,
-    execute_with_time_limit,
-    Benchmarker,
-)
 from unsloth_zoo.patching_utils import (
     patch_compiling_bitsandbytes,
     patch_layernorm,
@@ -127,8 +118,6 @@ from unsloth_zoo.gradient_checkpointing import (
     unsloth_offloaded_gradient_checkpoint,
     patch_unsloth_gradient_checkpointing,
     unpatch_unsloth_gradient_checkpointing,
-    Unsloth_Gradient_Checkpointer,
-    unsloth_gradient_checkpoint,
     patch_gradient_checkpointing,
     unpatch_gradient_checkpointing,
     patch_unsloth_smart_gradient_checkpointing,
@@ -313,7 +302,7 @@ class HideLoggingMessage(logging.Filter):
         self.text = text
 
     def filter(self, x):
-        return not (self.text in x.getMessage())
+        return self.text not in x.getMessage()
 
 
 # Replace warning messages (analogous to HideLoggingMessage but for warnings.warn)
@@ -737,21 +726,21 @@ def patch_mistral_nemo_config(config):
 try:
     # Some Config files use layer_type_validation
     # for eg Gemma-2, so we must import it to stop errors.
-    from transformers.configuration_utils import layer_type_validation
+    pass
 except:
     pass
 
 try:
     # Transformers 5.0+ uses RotaryEmbeddingConfigMixin as a base class for configs
-    from transformers.modeling_rope_utils import RotaryEmbeddingConfigMixin
+    pass
 except:
     pass
 from transformers import __version__ as transformers_version
 
 try:
-    from transformers import PreTrainedConfig
+    pass
 except:
-    from transformers import PretrainedConfig
+    pass
 
 model_architectures = [
     "llama",
@@ -850,7 +839,7 @@ from transformers.utils import is_openai_available
 
 if is_openai_available():
     try:
-        from openai import OpenAI
+        pass
     except:
         print("Unsloth: OpenAI failed to import - ignoring for now.")
         import transformers.utils
@@ -862,9 +851,7 @@ if is_openai_available():
 
 # =============================================
 # Get Flash Attention v2 if Ampere (RTX 30xx, A100)
-import bitsandbytes as bnb
 
-from transformers import AutoTokenizer
 from transformers.utils.import_utils import _is_package_available
 
 SUPPORTS_BFLOAT16 = False
@@ -882,9 +869,9 @@ if DEVICE_TYPE == "cuda":
             try:
                 try:
                     # See https://github.com/unslothai/unsloth/issues/1437
-                    from flash_attn.flash_attn_interface import flash_attn_gpu
+                    pass
                 except:
-                    from flash_attn.flash_attn_interface import flash_attn_cuda
+                    pass
                 HAS_FLASH_ATTENTION = True
 
                 # Also check for softcapping
@@ -931,9 +918,9 @@ elif DEVICE_TYPE == "hip":
         try:
             try:
                 # See https://github.com/unslothai/unsloth/issues/1437
-                from flash_attn.flash_attn_interface import flash_attn_gpu
+                pass
             except:
-                from flash_attn.flash_attn_interface import flash_attn_cuda
+                pass
             HAS_FLASH_ATTENTION = True
 
             # Also check for softcapping
@@ -1286,7 +1273,7 @@ USE_MODELSCOPE = os.environ.get("UNSLOTH_USE_MODELSCOPE", "0") == "1"
 if USE_MODELSCOPE:
     if importlib.util.find_spec("modelscope") is None:
         raise ImportError(
-            f"You are using the modelscope hub, please install modelscope by `pip install modelscope -U`"
+            "You are using the modelscope hub, please install modelscope by `pip install modelscope -U`"
         )
 
 import socket
@@ -1304,13 +1291,10 @@ def has_internet(host = "8.8.8.8", port = 53, timeout = 3):
             return True
         finally:
             sock.close()
-    except socket.error as ex:
+    except socket.error:
         return False
 
 
-import psutil
-
-
 def _get_statistics(statistics = None, force_download = True):
     # We log some basic stats about which environment is being used.
     # We simply download a README.md file from HF - all data is made public.
@@ -1477,7 +1461,6 @@ def get_statistics(local_files_only = False):
 # Fixes Bitsandbytes to remove missing warnings
 from transformers.utils.quantization_config import (
     BitsAndBytesConfig,
-    QuantizationMethod,
 )
 
 BitsAndBytesConfig__init__ = inspect.getsource(BitsAndBytesConfig.__init__)
diff --git a/unsloth/models/cohere.py b/unsloth/models/cohere.py
index 4251f3acd9..2716f3fd23 100644
--- a/unsloth/models/cohere.py
+++ b/unsloth/models/cohere.py
@@ -13,7 +13,6 @@
 # limitations under the License.
 
 from .llama import *
-from ._utils import __version__
 from unsloth_zoo.hf_utils import dtype_from_config
 from unsloth_zoo.utils import _get_dtype, Version
 from ..utils.packing import get_packed_info_from_kwargs
@@ -30,9 +29,6 @@ try:
         CohereDecoderLayer,
         CohereModel,
         CohereForCausalLM,
-        CohereRotaryEmbedding,
-        apply_rotary_pos_emb,
-        repeat_kv,
     )
 except:
     transformers_version = Version(transformers_version)
diff --git a/unsloth/models/falcon_h1.py b/unsloth/models/falcon_h1.py
index 6e3b16b21b..2fca99c088 100644
--- a/unsloth/models/falcon_h1.py
+++ b/unsloth/models/falcon_h1.py
@@ -13,8 +13,6 @@
 # limitations under the License.
 
 from .llama import *
-import os
-from ._utils import __version__
 from unsloth_zoo.utils import Version, _get_dtype
 from unsloth_zoo.hf_utils import dtype_from_config
 from ..utils.packing import get_packed_info_from_kwargs
@@ -28,7 +26,6 @@ from ..utils.attention_dispatch import (
 from .llama import (
     LlamaRotaryEmbedding,
     LlamaLinearScalingRotaryEmbedding,
-    _LlamaModel_fast_forward_inference,
 )
 
 try:
@@ -37,7 +34,6 @@ try:
         FalconH1DecoderLayer,
         FalconH1Model,
         FalconH1ForCausalLM,
-        FalconHybridMambaAttentionDynamicCache,
     )
 except:
     from transformers import __version__ as transformers_version
diff --git a/unsloth/models/gemma.py b/unsloth/models/gemma.py
index cf543ae094..c00596746f 100644
--- a/unsloth/models/gemma.py
+++ b/unsloth/models/gemma.py
@@ -14,14 +14,8 @@
 
 from .llama import *
 from .llama import _get_rope_theta
-from ._utils import __version__
 from unsloth_zoo.utils import _get_dtype, Version
 from unsloth_zoo.hf_utils import dtype_from_config
-from ..utils.packing import (
-    build_sdpa_packed_attention_mask,
-    build_xformers_block_causal_mask,
-    get_packed_info_from_kwargs,
-)
 import math
 
 try:
@@ -30,9 +24,6 @@ try:
         GemmaDecoderLayer,
         GemmaModel,
         GemmaForCausalLM,
-        GemmaRotaryEmbedding,
-        apply_rotary_pos_emb,
-        repeat_kv,
     )
 except:
     transformers_version = Version(transformers_version)
diff --git a/unsloth/models/gemma2.py b/unsloth/models/gemma2.py
index e59b8d5ebd..2aea84f751 100644
--- a/unsloth/models/gemma2.py
+++ b/unsloth/models/gemma2.py
@@ -13,7 +13,6 @@
 # limitations under the License.
 
 from .llama import *
-from ._utils import __version__
 from unsloth_zoo.utils import _get_dtype, Version
 from unsloth_zoo.hf_utils import dtype_from_config
 from ..utils.packing import get_packed_info_from_kwargs
@@ -22,7 +21,6 @@ from ..utils.attention_dispatch import (
     AttentionContext,
     run_attention,
     select_attention_backend,
-    SDPA,
 )
 from .gemma import (
     GemmaFixedRotaryEmbedding,
@@ -36,9 +34,6 @@ try:
         Gemma2DecoderLayer,
         Gemma2Model,
         Gemma2ForCausalLM,
-        Gemma2RotaryEmbedding,
-        apply_rotary_pos_emb,
-        repeat_kv,
     )
 except:
     transformers_version = Version(transformers_version)
@@ -65,7 +60,7 @@ except:
     Gemma2FlashAttention2 = Gemma2Attention
 
 if HAS_FLASH_ATTENTION_SOFTCAPPING:
-    from flash_attn import flash_attn_func
+    pass
 
 
 # Logit softcapping
diff --git a/unsloth/models/glm4_moe.py b/unsloth/models/glm4_moe.py
index 5d04b2f1d0..05a7d77e79 100644
--- a/unsloth/models/glm4_moe.py
+++ b/unsloth/models/glm4_moe.py
@@ -25,20 +25,11 @@ Key architecture differences from Qwen3 MoE:
 
 from .llama import *
 import os
-from ._utils import __version__
 from .llama import (
-    LlamaRotaryEmbedding,
-    LlamaLinearScalingRotaryEmbedding,
-    fix_prepare_inputs_for_generation,
     fast_rms_layernorm_inference,
     fast_swiglu_inference,
-    LlamaModel_fast_forward,
-    LlamaModel_fast_forward_inference,
-    CausalLM_fast_forward,
-    PeftModel_fast_forward,
 )
 import torch
-import torch.nn.functional as F
 from typing import Optional, Tuple
 from ..kernels import fast_rms_layernorm
 
diff --git a/unsloth/models/granite.py b/unsloth/models/granite.py
index 79ac41c43f..d522129900 100644
--- a/unsloth/models/granite.py
+++ b/unsloth/models/granite.py
@@ -13,7 +13,6 @@
 # limitations under the License.
 
 from .llama import *
-import os
 from ._utils import __version__
 from unsloth_zoo.utils import _get_dtype, Version
 from unsloth_zoo.hf_utils import dtype_from_config
@@ -265,8 +264,6 @@ def GraniteDecoderLayer_fast_forward(
     return outputs
 
 
-from math import sqrt as math_sqrt
-
 KV_CACHE_INCREMENT = 256  # KV Cache update size
 torch_nn_functional_softmax = torch.nn.functional.softmax
 torch_matmul = torch.matmul
@@ -285,7 +282,7 @@ def GraniteAttention_fast_forward_inference(
 ):
     assert (
         position_embeddings is not None
-    ), f"Granite model requires position embeddings to be specified"
+    ), "Granite model requires position embeddings to be specified"
 
     Xn = hidden_states
     bsz, _, hd = hidden_states.size()
diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 93d93e26d6..be5d42f627 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -26,7 +26,6 @@ from ._utils import (
     _get_inference_mode_context_manager,
     _prepare_model_for_qat,
     is_bfloat16_supported,
-    get_quant_type,
 )
 from .loader_utils import _get_fp8_mode_and_check_settings
 from ..utils.packing import (
@@ -50,12 +49,9 @@ from unsloth_zoo.hf_utils import (
 )
 from unsloth_zoo.peft_utils import SKIP_QUANTIZATION_MODULES
 from ..device_type import (
-    is_hip,
-    get_device_type,
     DEVICE_TYPE,
     DEVICE_TYPE_TORCH,
     DEVICE_COUNT,
-    ALLOW_PREQUANTIZED_MODELS,
 )
 
 transformers_version = Version(transformers_version)
@@ -97,7 +93,6 @@ except:
     LlamaFlashAttention2 = LlamaAttention
 
 from transformers import (
-    AutoTokenizer,
     AutoModelForCausalLM,
     AutoModelForSequenceClassification,
     BitsAndBytesConfig,
@@ -108,14 +103,16 @@ from transformers import set_seed as transformers_set_seed
 from peft import LoraConfig, TaskType, get_peft_model as _get_peft_model
 from peft import PeftModelForCausalLM, PeftModelForSequenceClassification
 from ..save import patch_saving_functions
-import re, os, inspect, math, sys
+import re
+import os
+import inspect
 import types
 
 try:
-    from huggingface_hub.utils import get_token
+    pass
 except:
     # Old HF Hub versions <= 0.0.25
-    from huggingface_hub.utils._token import get_token
+    pass
 from triton import __version__ as triton_version
 
 HAS_XFORMERS = xformers is not None
@@ -2983,7 +2980,7 @@ class FastLlamaModel:
                 try:
                     assert module in accepted_modules
                     final_modules.append(module)
-                except AssertionError as e:
+                except AssertionError:
                     final_modules.append(module)
                     print(
                         "Unsloth: You added custom modules, but Unsloth hasn't optimized for this.\n"
diff --git a/unsloth/models/loader.py b/unsloth/models/loader.py
index b54ceaf842..146f4a6b6d 100644
--- a/unsloth/models/loader.py
+++ b/unsloth/models/loader.py
@@ -15,20 +15,17 @@
 from ._utils import (
     _prepare_model_for_qat,
     is_bfloat16_supported,
-    is_vLLM_available,
     HAS_FLASH_ATTENTION,
     HAS_FLASH_ATTENTION_SOFTCAPPING,
     USE_MODELSCOPE,
     get_transformers_model_type,
     hf_login,
 )
-from .granite import FastGraniteModel
 from .llama import FastLlamaModel, logger
 from .mistral import FastMistralModel
 from .qwen2 import FastQwen2Model
 from .qwen3 import FastQwen3Model
 from .qwen3_moe import FastQwen3MoeModel
-from .cohere import FastCohereModel
 from transformers import AutoConfig
 from transformers import __version__ as transformers_version
 from peft import PeftConfig, PeftModel
@@ -39,22 +36,20 @@ from .loader_utils import (
     get_model_name,
     prepare_device_map,
 )
-import os, contextlib, sys
+import os
+import contextlib
 
 try:
-    from huggingface_hub import get_token
+    pass
 except:
     try:
-        from huggingface_hub.utils import get_token
+        pass
     except:
         # For older versions of huggingface_hub
-        from huggingface_hub.utils._token import get_token
+        pass
 from huggingface_hub import HfFileSystem
 import importlib.util
 from ..device_type import (
-    is_hip,
-    get_device_type,
-    DEVICE_TYPE,
     DEVICE_TYPE_TORCH,
     DEVICE_COUNT,
     ALLOW_PREQUANTIZED_MODELS,
@@ -86,15 +81,12 @@ if SUPPORTS_GEMMA:
 if SUPPORTS_GEMMA2:
     from .gemma2 import FastGemma2Model
 if SUPPORTS_FALCON_H1:
-    from .falcon_h1 import FastFalconH1Model
+    pass
 import torch
 from ._utils import (
     patch_compiling_bitsandbytes,
-    patch_model_and_tokenizer,
-    prepare_model_for_kbit_training,
     apply_unsloth_gradient_checkpointing,
     patch_compiled_autograd,
-    process_vision_info,
     unsloth_compile_transformers,
     fast_inference_setup,
 )
@@ -806,7 +798,6 @@ class FastLanguageModel(FastLlamaModel):
 
 from ..kernels import (
     patch_loss_functions,
-    post_patch_loss_function,
 )
 from .vision import FastBaseModel
 from transformers import (
diff --git a/unsloth/models/loader_utils.py b/unsloth/models/loader_utils.py
index cf5af983a6..9f571a2ae6 100644
--- a/unsloth/models/loader_utils.py
+++ b/unsloth/models/loader_utils.py
@@ -16,7 +16,6 @@ from ..device_type import DEVICE_TYPE_TORCH
 import importlib
 import os
 import torch
-import re
 import tempfile
 from typing import Union
 from .mapper import (
diff --git a/unsloth/models/mapper.py b/unsloth/models/mapper.py
index f0f430eb7e..38a289f551 100644
--- a/unsloth/models/mapper.py
+++ b/unsloth/models/mapper.py
@@ -20,1333 +20,1310 @@ __all__ = [
     "FLOAT_TO_FP8_ROW_MAPPER",
 ]
 
-__INT_TO_FLOAT_MAPPER = \
-{
-    "unsloth/mistral-7b-bnb-4bit" : (
+__INT_TO_FLOAT_MAPPER = {
+    "unsloth/mistral-7b-bnb-4bit": (
         "unsloth/mistral-7b",
         "mistralai/Mistral-7B-v0.1",
     ),
-    "unsloth/llama-2-7b-bnb-4bit" : (
+    "unsloth/llama-2-7b-bnb-4bit": (
         "unsloth/llama-2-7b",
         "meta-llama/Llama-2-7b-hf",
     ),
-    "unsloth/llama-2-13b-bnb-4bit" : (
+    "unsloth/llama-2-13b-bnb-4bit": (
         "unsloth/llama-2-13b",
         "meta-llama/Llama-2-13b-hf",
     ),
-    "unsloth/codellama-34b-bnb-4bit" : (
-        "codellama/CodeLlama-34b-hf",
-    ),
-    "unsloth/zephyr-sft-bnb-4bit" : (
+    "unsloth/codellama-34b-bnb-4bit": ("codellama/CodeLlama-34b-hf",),
+    "unsloth/zephyr-sft-bnb-4bit": (
         "unsloth/zephyr-sft",
         "HuggingFaceH4/mistral-7b-sft-beta",
     ),
-    "unsloth/tinyllama-bnb-4bit" : (
+    "unsloth/tinyllama-bnb-4bit": (
         "unsloth/tinyllama",
         "TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T",
     ),
-    "unsloth/tinyllama-chat-bnb-4bit" : (
+    "unsloth/tinyllama-chat-bnb-4bit": (
         "unsloth/tinyllama-chat",
         "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
     ),
-    "unsloth/mistral-7b-instruct-v0.1-bnb-4bit" : (
+    "unsloth/mistral-7b-instruct-v0.1-bnb-4bit": (
         "unsloth/mistral-7b-instruct-v0.1",
         "mistralai/Mistral-7B-Instruct-v0.1",
     ),
-    "unsloth/mistral-7b-instruct-v0.2-bnb-4bit" : (
+    "unsloth/mistral-7b-instruct-v0.2-bnb-4bit": (
         "unsloth/mistral-7b-instruct-v0.2",
         "mistralai/Mistral-7B-Instruct-v0.2",
     ),
-    "unsloth/llama-2-7b-chat-bnb-4bit" : (
+    "unsloth/llama-2-7b-chat-bnb-4bit": (
         "unsloth/llama-2-7b-chat",
         "meta-llama/Llama-2-7b-chat-hf",
     ),
-    "unsloth/llama-2-7b-chat-bnb-4bit" : (
+    "unsloth/llama-2-7b-chat-bnb-4bit": (
         "unsloth/llama-2-7b-chat",
         "meta-llama/Llama-2-7b-chat-hf",
     ),
-    "unsloth/Mixtral-8x7B-v0.1-unsloth-bnb-4bit" : (
+    "unsloth/Mixtral-8x7B-v0.1-unsloth-bnb-4bit": (
         "unsloth/Mixtral-8x7B-v0.1",
         "mistralai/Mixtral-8x7B-v0.1",
         "unsloth/Mixtral-8x7B-v0.1-bnb-4bit",
     ),
-    "unsloth/Mixtral-8x7B-Instruct-v0.1-unsloth-bnb-4bit" : (
+    "unsloth/Mixtral-8x7B-Instruct-v0.1-unsloth-bnb-4bit": (
         "unsloth/Mixtral-8x7B-Instruct-v0.1",
         "mistralai/Mixtral-8x7B-Instruct-v0.1",
         "unsloth/Mixtral-8x7B-Instruct-v0.1-bnb-4bit",
     ),
-    "unsloth/codellama-7b-bnb-4bit" : (
+    "unsloth/codellama-7b-bnb-4bit": (
         "unsloth/codellama-7b",
         "codellama/CodeLlama-7b-hf",
     ),
-    "unsloth/codellama-13b-bnb-4bit" : (
-        "codellama/CodeLlama-13b-hf",
-    ),
-    "unsloth/yi-6b-bnb-4bit" : (
+    "unsloth/codellama-13b-bnb-4bit": ("codellama/CodeLlama-13b-hf",),
+    "unsloth/yi-6b-bnb-4bit": (
         "unsloth/yi-6b",
         "01-ai/Yi-6B",
     ),
-    "unsloth/solar-10.7b-bnb-4bit" : (
-        "upstage/SOLAR-10.7B-v1.0",
-    ),
-    "unsloth/gemma-7b-bnb-4bit" : (
+    "unsloth/solar-10.7b-bnb-4bit": ("upstage/SOLAR-10.7B-v1.0",),
+    "unsloth/gemma-7b-bnb-4bit": (
         "unsloth/gemma-7b",
         "google/gemma-7b",
     ),
-    "unsloth/gemma-2b-bnb-4bit" : (
+    "unsloth/gemma-2b-bnb-4bit": (
         "unsloth/gemma-2b",
         "google/gemma-2b",
     ),
-    "unsloth/gemma-7b-it-bnb-4bit" : (
+    "unsloth/gemma-7b-it-bnb-4bit": (
         "unsloth/gemma-7b-it",
         "google/gemma-7b-it",
     ),
-    "unsloth/gemma-2b-bnb-4bit" : (
+    "unsloth/gemma-2b-bnb-4bit": (
         "unsloth/gemma-2b-it",
         "google/gemma-2b-it",
     ),
-    "unsloth/mistral-7b-v0.2-bnb-4bit" : (
+    "unsloth/mistral-7b-v0.2-bnb-4bit": (
         "unsloth/mistral-7b-v0.2",
         "alpindale/Mistral-7B-v0.2-hf",
     ),
-    "unsloth/gemma-1.1-2b-it-bnb-4bit" : (
+    "unsloth/gemma-1.1-2b-it-bnb-4bit": (
         "unsloth/gemma-1.1-2b-it",
         "google/gemma-1.1-2b-it",
     ),
-    "unsloth/gemma-1.1-7b-it-bnb-4bit" : (
+    "unsloth/gemma-1.1-7b-it-bnb-4bit": (
         "unsloth/gemma-1.1-7b-it",
         "google/gemma-1.1-7b-it",
     ),
-    "unsloth/Starling-LM-7B-beta" : (
+    "unsloth/Starling-LM-7B-beta": (
         "unsloth/Starling-LM-7B-beta",
         "Nexusflow/Starling-LM-7B-beta",
     ),
-    "unsloth/Hermes-2-Pro-Mistral-7B-bnb-4bit" : (
+    "unsloth/Hermes-2-Pro-Mistral-7B-bnb-4bit": (
         "unsloth/Hermes-2-Pro-Mistral-7B",
         "NousResearch/Hermes-2-Pro-Mistral-7B",
     ),
-    "unsloth/OpenHermes-2.5-Mistral-7B-bnb-4bit" : (
+    "unsloth/OpenHermes-2.5-Mistral-7B-bnb-4bit": (
         "unsloth/OpenHermes-2.5-Mistral-7B",
         "teknium/OpenHermes-2.5-Mistral-7B",
     ),
-    "unsloth/codegemma-2b-bnb-4bit" : (
+    "unsloth/codegemma-2b-bnb-4bit": (
         "unsloth/codegemma-2b",
         "google/codegemma-2b",
     ),
-    "unsloth/codegemma-7b-bnb-4bit" : (
+    "unsloth/codegemma-7b-bnb-4bit": (
         "unsloth/codegemma-7b",
         "google/codegemma-7b",
     ),
-    "unsloth/codegemma-7b-it-bnb-4bit" : (
+    "unsloth/codegemma-7b-it-bnb-4bit": (
         "unsloth/codegemma-7b-it",
         "google/codegemma-7b-it",
     ),
-    "unsloth/llama-3-8b-bnb-4bit" : (
+    "unsloth/llama-3-8b-bnb-4bit": (
         "unsloth/llama-3-8b",
         "meta-llama/Meta-Llama-3-8B",
     ),
-    "unsloth/llama-3-8b-Instruct-bnb-4bit" : (
+    "unsloth/llama-3-8b-Instruct-bnb-4bit": (
         "unsloth/llama-3-8b-Instruct",
         "meta-llama/Meta-Llama-3-8B-Instruct",
     ),
-    "unsloth/llama-3-70b-bnb-4bit" : (
-        "meta-llama/Meta-Llama-3-70B",
-    ),
-    "unsloth/llama-3-70b-Instruct-bnb-4bit" : (
-        "meta-llama/Meta-Llama-3-70B-Instruct",
-    ),
-    "unsloth/Phi-3-mini-4k-instruct-bnb-4bit" : (
+    "unsloth/llama-3-70b-bnb-4bit": ("meta-llama/Meta-Llama-3-70B",),
+    "unsloth/llama-3-70b-Instruct-bnb-4bit": ("meta-llama/Meta-Llama-3-70B-Instruct",),
+    "unsloth/Phi-3-mini-4k-instruct-bnb-4bit": (
         "unsloth/Phi-3-mini-4k-instruct",
         "microsoft/Phi-3-mini-4k-instruct",
     ),
-    "unsloth/mistral-7b-v0.3-bnb-4bit" : (
+    "unsloth/mistral-7b-v0.3-bnb-4bit": (
         "unsloth/mistral-7b-v0.3",
         "mistralai/Mistral-7B-v0.3",
     ),
-    "unsloth/mistral-7b-instruct-v0.3-bnb-4bit" : (
+    "unsloth/mistral-7b-instruct-v0.3-bnb-4bit": (
         "unsloth/mistral-7b-instruct-v0.3",
         "mistralai/Mistral-7B-Instruct-v0.3",
     ),
-    "unsloth/Phi-3-medium-4k-instruct-bnb-4bit" : (
+    "unsloth/Phi-3-medium-4k-instruct-bnb-4bit": (
         "unsloth/Phi-3-medium-4k-instruct",
         "microsoft/Phi-3-medium-4k-instruct",
     ),
-    "unsloth/Qwen2-0.5B-bnb-4bit" : (
+    "unsloth/Qwen2-0.5B-bnb-4bit": (
         "unsloth/Qwen2-0.5B",
         "Qwen/Qwen2-0.5B",
     ),
-    "unsloth/Qwen2-0.5B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2-0.5B-Instruct-bnb-4bit": (
         "unsloth/Qwen2-0.5B-Instruct",
         "Qwen/Qwen2-0.5B-Instruct",
     ),
-    "unsloth/Qwen2-1.5B-bnb-4bit" : (
+    "unsloth/Qwen2-1.5B-bnb-4bit": (
         "unsloth/Qwen2-1.5B",
         "Qwen/Qwen2-1.5B",
     ),
-    "unsloth/Qwen2-1.5B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2-1.5B-Instruct-bnb-4bit": (
         "unsloth/Qwen2-1.5B-Instruct",
         "Qwen/Qwen2-1.5B-Instruct",
     ),
-    "unsloth/Qwen2-7B-bnb-4bit" : (
+    "unsloth/Qwen2-7B-bnb-4bit": (
         "unsloth/Qwen2-7B",
         "Qwen/Qwen2-7B",
     ),
-    "unsloth/Qwen2-7B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2-7B-Instruct-bnb-4bit": (
         "unsloth/Qwen2-7B-Instruct",
         "Qwen/Qwen2-7B-Instruct",
     ),
-    "unsloth/Qwen2-70B-bnb-4bit" : (
-        "Qwen/Qwen2-70B",
-    ),
-    "unsloth/Qwen2-70B-Instruct-bnb-4bit" : (
-        "Qwen/Qwen2-70B-Instruct",
-    ),
-    "mistralai/Codestral-22B-v0.1" : (
-        "mistral-community/Codestral-22B-v0.1",
-    ),
-    "unsloth/gemma-2-9b-bnb-4bit" : (
+    "unsloth/Qwen2-70B-bnb-4bit": ("Qwen/Qwen2-70B",),
+    "unsloth/Qwen2-70B-Instruct-bnb-4bit": ("Qwen/Qwen2-70B-Instruct",),
+    "mistralai/Codestral-22B-v0.1": ("mistral-community/Codestral-22B-v0.1",),
+    "unsloth/gemma-2-9b-bnb-4bit": (
         "unsloth/gemma-2-9b",
         "google/gemma-2-9b",
     ),
-    "unsloth/gemma-2-27b-bnb-4bit" : (
+    "unsloth/gemma-2-27b-bnb-4bit": (
         "unsloth/gemma-2-27b",
         "google/gemma-2-27b",
     ),
-    "unsloth/gemma-2-9b-it-bnb-4bit" : (
+    "unsloth/gemma-2-9b-it-bnb-4bit": (
         "unsloth/gemma-2-9b-it",
         "google/gemma-2-9b-it",
     ),
-    "unsloth/gemma-2-27b-it-bnb-4bit" : (
+    "unsloth/gemma-2-27b-it-bnb-4bit": (
         "unsloth/gemma-2-27b-it",
         "google/gemma-2-27b-it",
     ),
-    "unsloth/Phi-3-mini-4k-instruct-v0-bnb-4bit" : ( # Old Phi pre July
+    "unsloth/Phi-3-mini-4k-instruct-v0-bnb-4bit": (  # Old Phi pre July
         "unsloth/Phi-3-mini-4k-instruct-v0",
     ),
-    "unsloth/Mistral-Nemo-Instruct-2407-bnb-4bit" : ( # New 12b Mistral models
+    "unsloth/Mistral-Nemo-Instruct-2407-bnb-4bit": (  # New 12b Mistral models
         "unsloth/Mistral-Nemo-Instruct-2407",
         "mistralai/Mistral-Nemo-Instruct-2407",
     ),
-    "unsloth/Mistral-Nemo-Base-2407-bnb-4bit" : ( # New 12b Mistral models
+    "unsloth/Mistral-Nemo-Base-2407-bnb-4bit": (  # New 12b Mistral models
         "unsloth/Mistral-Nemo-Base-2407",
         "mistralai/Mistral-Nemo-Base-2407",
     ),
-    "unsloth/Meta-Llama-3.1-8B-unsloth-bnb-4bit" : (
+    "unsloth/Meta-Llama-3.1-8B-unsloth-bnb-4bit": (
         "unsloth/Meta-Llama-3.1-8B",
         "meta-llama/Meta-Llama-3.1-8B",
         "unsloth/Meta-Llama-3.1-8B-bnb-4bit",
     ),
-    "unsloth/Meta-Llama-3.1-8B-Instruct-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Meta-Llama-3.1-8B-Instruct-unsloth-bnb-4bit": {
+        "8": (
             "RedHatAI/Llama-3.1-8B-Instruct-FP8",
             "unsloth/Llama-3.1-8B-Instruct-FP8-Block",
             "unsloth/Llama-3.1-8B-Instruct-FP8-Dynamic",
         ),
-        "16" : (
+        "16": (
             "unsloth/Meta-Llama-3.1-8B-Instruct",
             "meta-llama/Meta-Llama-3.1-8B-Instruct",
             "unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit",
         ),
     },
-    "unsloth/Llama-3.1-8B-unsloth-bnb-4bit" : (
+    "unsloth/Llama-3.1-8B-unsloth-bnb-4bit": (
         "unsloth/Llama-3.1-8B",
         "meta-llama/Llama-3.1-8B",
         "unsloth/Llama-3.1-8B-bnb-4bit",
     ),
-    "unsloth/Llama-3.1-8B-Instruct-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Llama-3.1-8B-Instruct-unsloth-bnb-4bit": {
+        "8": (
             "RedHatAI/Llama-3.1-8B-Instruct-FP8",
             "unsloth/Llama-3.1-8B-Instruct-FP8-Block",
             "unsloth/Llama-3.1-8B-Instruct-FP8-Dynamic",
         ),
-        "16" : (
+        "16": (
             "unsloth/Llama-3.1-8B-Instruct",
             "meta-llama/Llama-3.1-8B-Instruct",
             "unsloth/Llama-3.1-8B-Instruct-bnb-4bit",
         ),
     },
-    "unsloth/Meta-Llama-3.1-70B-bnb-4bit" : (
+    "unsloth/Meta-Llama-3.1-70B-bnb-4bit": (
         "unsloth/Meta-Llama-3.1-70B",
         "meta-llama/Meta-Llama-3.1-70B",
     ),
-    "unsloth/Meta-Llama-3.1-405B-bnb-4bit" : (
-        "meta-llama/Meta-Llama-3.1-405B",
-    ),
-    "unsloth/Meta-Llama-3.1-405B-Instruct-bnb-4bit" : (
+    "unsloth/Meta-Llama-3.1-405B-bnb-4bit": ("meta-llama/Meta-Llama-3.1-405B",),
+    "unsloth/Meta-Llama-3.1-405B-Instruct-bnb-4bit": (
         "meta-llama/Meta-Llama-3.1-405B-Instruct",
     ),
-    "unsloth/Meta-Llama-3.1-70B-Instruct-bnb-4bit" : (
+    "unsloth/Meta-Llama-3.1-70B-Instruct-bnb-4bit": (
         "unsloth/Meta-Llama-3.1-70B-Instruct",
         "meta-llama/Meta-Llama-3.1-70B-Instruct",
     ),
-    "unsloth/Mistral-Large-Instruct-2407-bnb-4bit" : (
+    "unsloth/Mistral-Large-Instruct-2407-bnb-4bit": (
         "mistralai/Mistral-Large-Instruct-2407",
     ),
-    "unsloth/gemma-2-2b-bnb-4bit" : (
+    "unsloth/gemma-2-2b-bnb-4bit": (
         "unsloth/gemma-2-2b",
         "google/gemma-2-2b",
     ),
-    "unsloth/gemma-2-2b-it-bnb-4bit" : (
+    "unsloth/gemma-2-2b-it-bnb-4bit": (
         "unsloth/gemma-2-2b-it",
         "google/gemma-2-2b-it",
     ),
-    "unsloth/Phi-3.5-mini-instruct-bnb-4bit" : (
+    "unsloth/Phi-3.5-mini-instruct-bnb-4bit": (
         "unsloth/Phi-3.5-mini-instruct",
         "microsoft/Phi-3.5-mini-instruct",
     ),
-    "unsloth/c4ai-command-r-08-2024-bnb-4bit" : (
-        "CohereForAI/c4ai-command-r-08-2024",
-    ),
-    "unsloth/c4ai-command-r-plus-08-2024-bnb-4bit" : (
+    "unsloth/c4ai-command-r-08-2024-bnb-4bit": ("CohereForAI/c4ai-command-r-08-2024",),
+    "unsloth/c4ai-command-r-plus-08-2024-bnb-4bit": (
         "CohereForAI/c4ai-command-r-plus-08-2024",
     ),
-    "unsloth/Llama-3.1-Storm-8B-bnb-4bit" : (
+    "unsloth/Llama-3.1-Storm-8B-bnb-4bit": (
         "unsloth/Llama-3.1-Storm-8B",
         "akjindal53244/Llama-3.1-Storm-8B",
     ),
-    "unsloth/Hermes-3-Llama-3.1-8B-bnb-4bit" : (
+    "unsloth/Hermes-3-Llama-3.1-8B-bnb-4bit": (
         "unsloth/Hermes-3-Llama-3.1-8B",
         "NousResearch/Hermes-3-Llama-3.1-8B",
     ),
-    "unsloth/Hermes-3-Llama-3.1-70B-bnb-4bit" : (
+    "unsloth/Hermes-3-Llama-3.1-70B-bnb-4bit": (
         "unsloth/Hermes-3-Llama-3.1-70B",
         "NousResearch/Hermes-3-Llama-3.1-70B",
     ),
-    "unsloth/Hermes-3-Llama-3.1-405B-bnb-4bit" : (
+    "unsloth/Hermes-3-Llama-3.1-405B-bnb-4bit": (
         "NousResearch/Hermes-3-Llama-3.1-405B",
     ),
-    "unsloth/SmolLM-135M-bnb-4bit" : (
+    "unsloth/SmolLM-135M-bnb-4bit": (
         "unsloth/SmolLM-135M",
         "HuggingFaceTB/SmolLM-135M",
     ),
-    "unsloth/SmolLM-360M-bnb-4bit" : (
+    "unsloth/SmolLM-360M-bnb-4bit": (
         "unsloth/SmolLM-360M",
         "HuggingFaceTB/SmolLM-360M",
     ),
-    "unsloth/SmolLM-1.7B-bnb-4bit" : (
+    "unsloth/SmolLM-1.7B-bnb-4bit": (
         "unsloth/SmolLM-1.7B",
         "HuggingFaceTB/SmolLM-1.7B",
     ),
-    "unsloth/SmolLM-135M-Instruct-bnb-4bit" : (
+    "unsloth/SmolLM-135M-Instruct-bnb-4bit": (
         "unsloth/SmolLM-135M-Instruct",
         "HuggingFaceTB/SmolLM-135M-Instruct",
     ),
-    "unsloth/SmolLM-360M-Instruct-bnb-4bit" : (
+    "unsloth/SmolLM-360M-Instruct-bnb-4bit": (
         "unsloth/SmolLM-360M-Instruct",
         "HuggingFaceTB/SmolLM-360M-Instruct",
     ),
-    "unsloth/SmolLM-1.7B-Instruct-bnb-4bit" : (
+    "unsloth/SmolLM-1.7B-Instruct-bnb-4bit": (
         "unsloth/SmolLM-1.7B-Instruct",
         "HuggingFaceTB/SmolLM-1.7B-Instruct",
     ),
-    "unsloth/Mistral-Small-Instruct-2409-bnb-4bit" : (
+    "unsloth/Mistral-Small-Instruct-2409-bnb-4bit": (
         "unsloth/Mistral-Small-Instruct-2409",
         "mistralai/Mistral-Small-Instruct-2409",
     ),
-    "unsloth/Qwen2.5-0.5B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-0.5B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-0.5B-Instruct",
         "Qwen/Qwen2.5-0.5B-Instruct",
         "unsloth/Qwen2.5-0.5B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-1.5B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-1.5B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-1.5B-Instruct",
         "Qwen/Qwen2.5-1.5B-Instruct",
         "unsloth/Qwen2.5-1.5B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-3B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-3B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-3B-Instruct",
         "Qwen/Qwen2.5-3B-Instruct",
         "unsloth/Qwen2.5-3B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-7B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-7B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-7B-Instruct",
         "Qwen/Qwen2.5-7B-Instruct",
         "unsloth/Qwen2.5-7B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-14B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-14B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-14B-Instruct",
         "Qwen/Qwen2.5-14B-Instruct",
         "unsloth/Qwen2.5-14B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-32B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-32B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-32B-Instruct",
         "Qwen/Qwen2.5-32B-Instruct",
     ),
-    "unsloth/Qwen2.5-72B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-72B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-72B-Instruct",
         "Qwen/Qwen2.5-72B-Instruct",
     ),
-    "unsloth/Qwen2.5-0.5B-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-0.5B-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-0.5B",
         "Qwen/Qwen2.5-0.5B",
         "unsloth/Qwen2.5-0.5B-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-1.5B-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-1.5B-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-1.5B",
         "Qwen/Qwen2.5-1.5B",
         "unsloth/Qwen2.5-1.5B-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-3B-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-3B-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-3B",
         "Qwen/Qwen2.5-3B",
         "unsloth/Qwen2.5-3B-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-7B-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-7B-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-7B",
         "Qwen/Qwen2.5-7B",
         "unsloth/Qwen2.5-7B-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-14B-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-14B-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-14B",
         "Qwen/Qwen2.5-14B",
         "unsloth/Qwen2.5-14B-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-32B-bnb-4bit" : (
+    "unsloth/Qwen2.5-32B-bnb-4bit": (
         "unsloth/Qwen2.5-32B",
         "Qwen/Qwen2.5-32B",
     ),
-    "unsloth/Qwen2.5-72B-bnb-4bit" : (
+    "unsloth/Qwen2.5-72B-bnb-4bit": (
         "unsloth/Qwen2.5-72B",
         "Qwen/Qwen2.5-72B",
     ),
-    "unsloth/Qwen2.5-Math-1.5B-bnb-4bit" : (
+    "unsloth/Qwen2.5-Math-1.5B-bnb-4bit": (
         "unsloth/Qwen2.5-Math-1.5B",
         "Qwen/Qwen2.5-Math-1.5B",
     ),
-    "unsloth/Qwen2.5-Math-7B-bnb-4bit" : (
+    "unsloth/Qwen2.5-Math-7B-bnb-4bit": (
         "unsloth/Qwen2.5-Math-7B",
         "Qwen/Qwen2.5-Math-7B",
     ),
-    "unsloth/Qwen2.5-Math-72B-bnb-4bit" : (
+    "unsloth/Qwen2.5-Math-72B-bnb-4bit": (
         "unsloth/Qwen2.5-Math-72B",
         "Qwen/Qwen2.5-Math-72B",
     ),
-    "unsloth/Qwen2.5-Math-1.5B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-Math-1.5B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-Math-1.5B-Instruct",
         "Qwen/Qwen2.5-Math-1.5B-Instruct",
     ),
-    "unsloth/Qwen2.5-Math-7B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-Math-7B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-Math-7B-Instruct",
         "Qwen/Qwen2.5-Math-7B-Instruct",
     ),
-    "unsloth/Qwen2.5-Math-72B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-Math-72B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-Math-72B-Instruct",
         "Qwen/Qwen2.5-Math-72B-Instruct",
     ),
-    "unsloth/Qwen2.5-Coder-0.5B-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-0.5B-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-0.5B",
         "Qwen/Qwen2.5-Coder-0.5B",
     ),
-    "unsloth/Qwen2.5-Coder-1.5B-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-1.5B-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-1.5B",
         "Qwen/Qwen2.5-Coder-1.5B",
     ),
-    "unsloth/Qwen2.5-Coder-3B-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-3B-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-3B",
         "Qwen/Qwen2.5-Coder-3B",
     ),
-    "unsloth/Qwen2.5-Coder-7B-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-7B-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-7B",
         "Qwen/Qwen2.5-Coder-7B",
     ),
-    "unsloth/Qwen2.5-Coder-14B-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-14B-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-14B",
         "Qwen/Qwen2.5-Coder-14B",
     ),
-    "unsloth/Qwen2.5-Coder-32B-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-32B-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-32B",
         "Qwen/Qwen2.5-Coder-32B",
     ),
-    "unsloth/Qwen2.5-Coder-0.5B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-0.5B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-0.5B-Instruct",
         "Qwen/Qwen2.5-Coder-0.5B-Instruct",
     ),
-    "unsloth/Qwen2.5-Coder-1.5B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-1.5B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-1.5B-Instruct",
         "Qwen/Qwen2.5-Coder-1.5B-Instruct",
     ),
-    "unsloth/Qwen2.5-Coder-3B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-3B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-3B-Instruct",
         "Qwen/Qwen2.5-Coder-3B-Instruct",
     ),
-    "unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-7B-Instruct",
         "Qwen/Qwen2.5-Coder-7B-Instruct",
     ),
-    "unsloth/Qwen2.5-Coder-14B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-14B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-14B-Instruct",
         "Qwen/Qwen2.5-Coder-14B-Instruct",
     ),
-    "unsloth/Qwen2.5-Coder-32B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2.5-Coder-32B-Instruct-bnb-4bit": (
         "unsloth/Qwen2.5-Coder-32B-Instruct",
         "Qwen/Qwen2.5-Coder-32B-Instruct",
     ),
-    "unsloth/Llama-3.2-1B-unsloth-bnb-4bit" : (
+    "unsloth/Llama-3.2-1B-unsloth-bnb-4bit": (
         "unsloth/Llama-3.2-1B",
         "meta-llama/Llama-3.2-1B",
         "unsloth/Llama-3.2-1B-bnb-4bit",
     ),
-    "unsloth/Llama-3.2-3B-unsloth-bnb-4bit" : (
+    "unsloth/Llama-3.2-3B-unsloth-bnb-4bit": (
         "unsloth/Llama-3.2-3B",
         "meta-llama/Llama-3.2-3B",
         "unsloth/Llama-3.2-3B-bnb-4bit",
     ),
-    "unsloth/Llama-3.2-1B-Instruct-unsloth-bnb-4bit" : {
+    "unsloth/Llama-3.2-1B-Instruct-unsloth-bnb-4bit": {
         "8": (
             "RedHatAI/Llama-3.2-1B-Instruct-FP8",
             "unsloth/Llama-3.2-1B-Instruct-FP8-Block",
             "unsloth/Llama-3.2-1B-Instruct-FP8-Dynamic",
         ),
-        "16" : (
+        "16": (
             "unsloth/Llama-3.2-1B-Instruct",
             "meta-llama/Llama-3.2-1B-Instruct",
             "unsloth/Llama-3.2-1B-Instruct-bnb-4bit",
         ),
     },
-    "unsloth/Llama-3.2-3B-Instruct-unsloth-bnb-4bit" : {
+    "unsloth/Llama-3.2-3B-Instruct-unsloth-bnb-4bit": {
         "8": (
             "RedHatAI/Llama-3.2-3B-Instruct-FP8",
             "unsloth/Llama-3.2-3B-Instruct-FP8-Block",
             "unsloth/Llama-3.2-3B-Instruct-FP8-Dynamic",
         ),
-        "16" : (
+        "16": (
             "unsloth/Llama-3.2-3B-Instruct",
             "meta-llama/Llama-3.2-3B-Instruct",
             "unsloth/Llama-3.2-3B-Instruct-bnb-4bit",
         ),
     },
-    "unsloth/Llama-3.1-Nemotron-70B-Instruct-bnb-4bit" : (
+    "unsloth/Llama-3.1-Nemotron-70B-Instruct-bnb-4bit": (
         "unsloth/Llama-3.1-Nemotron-70B-Instruct",
         "nvidia/Llama-3.1-Nemotron-70B-Instruct-HF",
     ),
-    "unsloth/Qwen2-VL-2B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2-VL-2B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2-VL-2B-Instruct",
         "Qwen/Qwen2-VL-2B-Instruct",
         "unsloth/Qwen2-VL-2B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2-VL-7B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2-VL-7B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2-VL-7B-Instruct",
         "Qwen/Qwen2-VL-7B-Instruct",
         "unsloth/Qwen2-VL-7B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2-VL-72B-Instruct-bnb-4bit" : (
+    "unsloth/Qwen2-VL-72B-Instruct-bnb-4bit": (
         "unsloth/Qwen2-VL-72B-Instruct",
         "Qwen/Qwen2-VL-72B-Instruct",
     ),
-    "unsloth/Qwen2-VL-2B-bnb-4bit" : (
+    "unsloth/Qwen2-VL-2B-bnb-4bit": (
         "unsloth/Qwen2-VL-2B",
         "Qwen/Qwen2-VL-2B",
     ),
-    "unsloth/Qwen2-VL-7B-bnb-4bit" : (
+    "unsloth/Qwen2-VL-7B-bnb-4bit": (
         "unsloth/Qwen2-VL-7B",
         "Qwen/Qwen2-VL-7B",
     ),
-    "unsloth/Qwen2-VL-72B-bnb-4bit" : (
+    "unsloth/Qwen2-VL-72B-bnb-4bit": (
         "unsloth/Qwen2-VL-72B",
         "Qwen/Qwen2-VL-72B",
     ),
-    "unsloth/Llama-3.2-11B-Vision-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Llama-3.2-11B-Vision-Instruct-unsloth-bnb-4bit": (
         "unsloth/Llama-3.2-11B-Vision-Instruct",
         "meta-llama/Llama-3.2-11B-Vision-Instruct",
         "unsloth/Llama-3.2-11B-Vision-Instruct-bnb-4bit",
     ),
-    "unsloth/Llama-3.2-90B-Vision-Instruct-bnb-4bit" : (
+    "unsloth/Llama-3.2-90B-Vision-Instruct-bnb-4bit": (
         "unsloth/Llama-3.2-90B-Vision-Instruct",
         "meta-llama/Llama-3.2-90B-Vision-Instruct",
     ),
-    "unsloth/Llama-3.2-11B-Vision-unsloth-bnb-4bit" : (
+    "unsloth/Llama-3.2-11B-Vision-unsloth-bnb-4bit": (
         "unsloth/Llama-3.2-11B-Vision",
         "meta-llama/Llama-3.2-11B-Vision",
         "unsloth/Llama-3.2-11B-Vision-bnb-4bit",
     ),
-    "unsloth/Llama-3.2-90B-Vision-bnb-4bit" : (
+    "unsloth/Llama-3.2-90B-Vision-bnb-4bit": (
         "unsloth/Llama-3.2-90B-Vision",
         "meta-llama/Llama-3.2-90B-Vision",
     ),
-    "unsloth/Pixtral-12B-2409-unsloth-bnb-4bit" : (
+    "unsloth/Pixtral-12B-2409-unsloth-bnb-4bit": (
         "unsloth/Pixtral-12B-2409",
         "mistralai/Pixtral-12B-2409",
         "unsloth/Pixtral-12B-2409-bnb-4bit",
     ),
-    "unsloth/Pixtral-12B-2409-Base-bnb-4bit" : (
+    "unsloth/Pixtral-12B-2409-Base-bnb-4bit": (
         "unsloth/Pixtral-12B-Base-2409",
         "mistralai/Pixtral-12B-Base-2409",
     ),
-    "unsloth/llava-1.5-7b-hf-bnb-4bit" : (
+    "unsloth/llava-1.5-7b-hf-bnb-4bit": (
         "unsloth/llava-1.5-7b-hf",
         "llava-hf/llava-1.5-7b-hf",
     ),
-    "unsloth/llava-v1.6-mistral-7b-hf-bnb-4bit" : (
+    "unsloth/llava-v1.6-mistral-7b-hf-bnb-4bit": (
         "unsloth/llava-v1.6-mistral-7b-hf",
         "llava-hf/llava-v1.6-mistral-7b-hf",
     ),
-    "unsloth/Llama-3.1-Tulu-3-8B-bnb-4bit" : (
+    "unsloth/Llama-3.1-Tulu-3-8B-bnb-4bit": (
         "unsloth/Llama-3.1-Tulu-3-8B",
         "allenai/Llama-3.1-Tulu-3-8B",
     ),
-    "unsloth/Llama-3.1-Tulu-3-70B-bnb-4bit" : (
+    "unsloth/Llama-3.1-Tulu-3-70B-bnb-4bit": (
         "unsloth/Llama-3.1-Tulu-3-70B",
         "allenai/Llama-3.1-Tulu-3-70B",
     ),
-    "unsloth/QwQ-32B-Preview-bnb-4bit" : (
+    "unsloth/QwQ-32B-Preview-bnb-4bit": (
         "unsloth/QwQ-32B-Preview",
         "Qwen/QwQ-32B-Preview",
     ),
-    "unsloth/Llama-3.3-70B-Instruct-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Llama-3.3-70B-Instruct-unsloth-bnb-4bit": {
+        "8": (
             "RedHatAI/Llama-3.3-70B-Instruct-FP8",
             "unsloth/Llama-3.3-70B-Instruct-FP8-Block",
             "unsloth/Llama-3.3-70B-Instruct-FP8-Dynamic",
         ),
-        "16" : (
+        "16": (
             "unsloth/Llama-3.3-70B-Instruct",
             "meta-llama/Llama-3.3-70B-Instruct",
             "unsloth/Llama-3.3-70B-Instruct-bnb-4bit",
         ),
     },
-    "unsloth/phi-4-unsloth-bnb-4bit" : (
+    "unsloth/phi-4-unsloth-bnb-4bit": (
         "unsloth/phi-4",
         "microsoft/phi-4",
         "unsloth/phi-4-bnb-4bit",
     ),
-    "unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit" : (
+    "unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit": (
         "unsloth/DeepSeek-R1-Distill-Qwen-32B",
         "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
     ),
-    "unsloth/DeepSeek-R1-Distill-Qwen-14B-unsloth-bnb-4bit" : (
+    "unsloth/DeepSeek-R1-Distill-Qwen-14B-unsloth-bnb-4bit": (
         "unsloth/DeepSeek-R1-Distill-Qwen-14B",
         "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B",
         "unsloth/DeepSeek-R1-Distill-Qwen-14B-bnb-4bit",
     ),
-    "unsloth/DeepSeek-R1-Distill-Qwen-7B-unsloth-bnb-4bit" : (
+    "unsloth/DeepSeek-R1-Distill-Qwen-7B-unsloth-bnb-4bit": (
         "unsloth/DeepSeek-R1-Distill-Qwen-7B",
         "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
         "unsloth/DeepSeek-R1-Distill-Qwen-7B-bnb-4bit",
     ),
-    "unsloth/DeepSeek-R1-Distill-Qwen-1.5B-unsloth-bnb-4bit" : (
+    "unsloth/DeepSeek-R1-Distill-Qwen-1.5B-unsloth-bnb-4bit": (
         "unsloth/DeepSeek-R1-Distill-Qwen-1.5B",
         "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
         "unsloth/DeepSeek-R1-Distill-Qwen-1.5B-bnb-4bit",
     ),
-    "unsloth/DeepSeek-R1-Distill-Llama-8B-unsloth-bnb-4bit" : (
+    "unsloth/DeepSeek-R1-Distill-Llama-8B-unsloth-bnb-4bit": (
         "unsloth/DeepSeek-R1-Distill-Llama-8B",
         "deepseek-ai/DeepSeek-R1-Distill-Llama-8B",
         "unsloth/DeepSeek-R1-Distill-Llama-8B-bnb-4bit",
     ),
-    "unsloth/DeepSeek-R1-Distill-Llama-70B-bnb-4bit" : (
+    "unsloth/DeepSeek-R1-Distill-Llama-70B-bnb-4bit": (
         "unsloth/DeepSeek-R1-Distill-Llama-70B",
         "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
     ),
-    "unsloth/Mistral-Small-24B-Base-2501-unsloth-bnb-4bit" : (
+    "unsloth/Mistral-Small-24B-Base-2501-unsloth-bnb-4bit": (
         "unsloth/Mistral-Small-24B-Base-2501",
         "mistralai/Mistral-Small-24B-Base-2501",
         "unsloth/Mistral-Small-24B-Base-2501-bnb-4bit",
     ),
-    "unsloth/Mistral-Small-24B-Instruct-2501-unsloth-bnb-4bit" : (
+    "unsloth/Mistral-Small-24B-Instruct-2501-unsloth-bnb-4bit": (
         "unsloth/Mistral-Small-24B-Instruct-2501",
         "mistralai/Mistral-Small-24B-Instruct-2501",
         "unsloth/Mistral-Small-24B-Instruct-2501-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-VL-3B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-VL-3B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-VL-3B-Instruct",
         "Qwen/Qwen2.5-VL-3B-Instruct",
         "unsloth/Qwen2.5-VL-3B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-VL-7B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-VL-7B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-VL-7B-Instruct",
         "Qwen/Qwen2.5-VL-7B-Instruct",
         "unsloth/Qwen2.5-VL-7B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-VL-32B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-VL-32B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-VL-32B-Instruct",
         "Qwen/Qwen2.5-VL-32B-Instruct",
         "unsloth/Qwen2.5-VL-32B-Instruct-bnb-4bit",
     ),
-    "unsloth/Qwen2.5-VL-72B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/Qwen2.5-VL-72B-Instruct-unsloth-bnb-4bit": (
         "unsloth/Qwen2.5-VL-72B-Instruct",
         "Qwen/Qwen2.5-VL-72B-Instruct",
         "unsloth/Qwen2.5-VL-72B-Instruct-bnb-4bit",
     ),
-    "unsloth/DeepScaleR-1.5B-Preview-unsloth-bnb-4bit" : (
+    "unsloth/DeepScaleR-1.5B-Preview-unsloth-bnb-4bit": (
         "unsloth/DeepHermes-3-Llama-3-8B-Preview",
         "agentica-org/DeepScaleR-1.5B-Preview",
         "unsloth/DeepScaleR-1.5B-Preview-bnb-4bit",
     ),
-    "unsloth/OpenThinker-7B-unsloth-bnb-4bit" : (
+    "unsloth/OpenThinker-7B-unsloth-bnb-4bit": (
         "unsloth/OpenThinker-7B",
         "open-thoughts/OpenThinker-7B",
         "unsloth/OpenThinker-7B-bnb-4bit",
     ),
-    "unsloth/granite-3.2-2b-instruct-unsloth-bnb-4bit" : (
+    "unsloth/granite-3.2-2b-instruct-unsloth-bnb-4bit": (
         "unsloth/granite-3.2-2b-instruct",
         "ibm-granite/granite-3.2-2b-instruct",
         "unsloth/granite-3.2-2b-instruct-bnb-4bit",
     ),
-    "unsloth/granite-3.2-8b-instruct-unsloth-bnb-4bit" : (
+    "unsloth/granite-3.2-8b-instruct-unsloth-bnb-4bit": (
         "unsloth/granite-3.2-8b-instruct",
         "ibm-granite/granite-3.2-8b-instruct",
         "unsloth/granite-3.2-8b-instruct-bnb-4bit",
     ),
-    "unsloth/QwQ-32B-unsloth-bnb-4bit" : (
+    "unsloth/QwQ-32B-unsloth-bnb-4bit": (
         "unsloth/QwQ-32B",
         "Qwen/QwQ-32B",
         "unsloth/QwQ-32B-bnb-4bit",
     ),
-    "unsloth/gemma-3-1b-it-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-1b-it-unsloth-bnb-4bit": (
         "unsloth/gemma-3-1b-it",
         "google/gemma-3-1b-it",
         "unsloth/gemma-3-1b-it-bnb-4bit",
     ),
-    "unsloth/gemma-3-4b-it-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-4b-it-unsloth-bnb-4bit": (
         "unsloth/gemma-3-4b-it",
         "google/gemma-3-4b-it",
         "unsloth/gemma-3-4b-it-bnb-4bit",
     ),
-    "unsloth/gemma-3-12b-it-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-12b-it-unsloth-bnb-4bit": (
         "unsloth/gemma-3-12b-it",
         "google/gemma-3-12b-it",
         "unsloth/gemma-3-12b-it-bnb-4bit",
     ),
-    "unsloth/gemma-3-27b-it-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-27b-it-unsloth-bnb-4bit": (
         "unsloth/gemma-3-27b-it",
         "google/gemma-3-27b-it",
         "unsloth/gemma-3-27b-it-bnb-4bit",
     ),
-    "unsloth/gemma-3-1b-pt-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-1b-pt-unsloth-bnb-4bit": (
         "unsloth/gemma-3-1b-pt",
         "google/gemma-3-1b-pt",
         "unsloth/gemma-3-1b-pt-bnb-4bit",
     ),
-    "unsloth/gemma-3-4b-pt-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-4b-pt-unsloth-bnb-4bit": (
         "unsloth/gemma-3-4b-pt",
         "google/gemma-3-4b-pt",
         "unsloth/gemma-3-4b-pt-bnb-4bit",
     ),
-    "unsloth/gemma-3-12b-pt-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-12b-pt-unsloth-bnb-4bit": (
         "unsloth/gemma-3-12b-pt",
         "google/gemma-3-12b-pt",
         "unsloth/gemma-3-12b-pt-bnb-4bit",
     ),
-    "unsloth/gemma-3-27b-pt-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-27b-pt-unsloth-bnb-4bit": (
         "unsloth/gemma-3-27b-pt",
         "google/gemma-3-27b-pt",
         "unsloth/gemma-3-27b-pt-bnb-4bit",
     ),
-    "unsloth/reka-flash-3-unsloth-bnb-4bit" : (
+    "unsloth/reka-flash-3-unsloth-bnb-4bit": (
         "unsloth/reka-flash-3",
         "RekaAI/reka-flash-3",
         "unsloth/reka-flash-3-bnb-4bit",
     ),
-    "unsloth/c4ai-command-a-03-2025-unsloth-bnb-4bit" : (
+    "unsloth/c4ai-command-a-03-2025-unsloth-bnb-4bit": (
         "unsloth/c4ai-command-a-03-2025",
         "CohereForAI/c4ai-command-a-03-2025",
         "unsloth/c4ai-command-a-03-2025-bnb-4bit",
     ),
-    "unsloth/aya-vision-32b-unsloth-bnb-4bit" : (
+    "unsloth/aya-vision-32b-unsloth-bnb-4bit": (
         "unsloth/aya-vision-32b",
         "CohereForAI/aya-vision-32b",
         "unsloth/aya-vision-32b-bnb-4bit",
     ),
-    "unsloth/aya-vision-8b-unsloth-bnb-4bit" : (
+    "unsloth/aya-vision-8b-unsloth-bnb-4bit": (
         "unsloth/aya-vision-8b",
         "CohereForAI/aya-vision-8b",
         "unsloth/aya-vision-8b-bnb-4bit",
     ),
-    "unsloth/granite-vision-3.2-2b-unsloth-bnb-4bit" : (
+    "unsloth/granite-vision-3.2-2b-unsloth-bnb-4bit": (
         "unsloth/granite-vision-3.2-2b",
         "ibm-granite/granite-vision-3.2-2b",
         "unsloth/granite-vision-3.2-2b-bnb-4bit",
     ),
-    "unsloth/OLMo-2-0325-32B-Instruct-unsloth-bnb-4bit" : (
+    "unsloth/OLMo-2-0325-32B-Instruct-unsloth-bnb-4bit": (
         "unsloth/OLMo-2-0325-32B-Instruct",
         "allenai/OLMo-2-0325-32B-Instruct",
         "unsloth/OLMo-2-0325-32B-Instruct-bnb-4bit",
     ),
-    "unsloth/Mistral-Small-3.1-24B-Instruct-2503-unsloth-bnb-4bit" : (
+    "unsloth/Mistral-Small-3.1-24B-Instruct-2503-unsloth-bnb-4bit": (
         "unsloth/Mistral-Small-3.1-24B-Instruct-2503",
         "mistralai/Mistral-Small-3.1-24B-Instruct-2503",
         "unsloth/Mistral-Small-3.1-24B-Instruct-2503-bnb-4bit",
     ),
-    "unsloth/Mistral-Small-3.1-24B-Base-2503-unsloth-bnb-4bit" : (
+    "unsloth/Mistral-Small-3.1-24B-Base-2503-unsloth-bnb-4bit": (
         "unsloth/Mistral-Small-3.1-24B-Base-2503",
         "mistralai/Mistral-Small-3.1-24B-Base-2503",
         "unsloth/Mistral-Small-3.1-24B-Base-2503-bnb-4bit",
     ),
-    "unsloth/Qwen3-0.6B-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-0.6B-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-0.6B-FP8",
             "unsloth/Qwen3-0.6B-FP8",
             "unsloth/Qwen3-0.6B-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-0.6B",
             "Qwen/Qwen3-0.6B",
             "unsloth/Qwen3-0.6B-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-1.7B-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-1.7B-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-1.7B-FP8",
             "unsloth/Qwen3-1.7B-FP8",
             "unsloth/Qwen3-1.7B-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-1.7B",
             "Qwen/Qwen3-1.7B",
             "unsloth/Qwen3-1.7B-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-4B-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-4B-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-4B-FP8",
             "unsloth/Qwen3-4B-FP8",
             "unsloth/Qwen3-4B-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-4B",
             "Qwen/Qwen3-4B",
             "unsloth/Qwen3-4B-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-8B-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-8B-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-8B-FP8",
             "unsloth/Qwen3-8B-FP8",
             "unsloth/Qwen3-8B-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-8B",
             "Qwen/Qwen3-8B",
             "unsloth/Qwen3-8B-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-14B-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-14B-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-14B-FP8",
             "unsloth/Qwen3-14B-FP8",
             "unsloth/Qwen3-14B-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-14B",
             "Qwen/Qwen3-14B",
             "unsloth/Qwen3-14B-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-32B-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-32B-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-32B-FP8",
             "unsloth/Qwen3-32B-FP8",
             "unsloth/Qwen3-32B-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-32B",
             "Qwen/Qwen3-32B",
             "unsloth/Qwen3-32B-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-30B-A3B-unsloth-bnb-4bit" : (
+    "unsloth/Qwen3-30B-A3B-unsloth-bnb-4bit": (
         "unsloth/Qwen3-30B-A3B",
         "Qwen/Qwen3-30B-A3B",
         "unsloth/Qwen3-30B-A3B-bnb-4bit",
     ),
-    "unsloth/Qwen3-0.6B-Base-unsloth-bnb-4bit" : (
+    "unsloth/Qwen3-0.6B-Base-unsloth-bnb-4bit": (
         "unsloth/Qwen3-0.6B-Base",
         "Qwen/Qwen3-0.6B-Base",
         "unsloth/Qwen3-0.6B-Base-bnb-4bit",
     ),
-    "unsloth/Qwen3-1.7B-Base-unsloth-bnb-4bit" : (
+    "unsloth/Qwen3-1.7B-Base-unsloth-bnb-4bit": (
         "unsloth/Qwen3-1.7B-Base",
         "Qwen/Qwen3-1.7B-Base",
         "unsloth/Qwen3-1.7B-Base-bnb-4bit",
     ),
-    "unsloth/Qwen3-4B-Base-unsloth-bnb-4bit" : (
+    "unsloth/Qwen3-4B-Base-unsloth-bnb-4bit": (
         "unsloth/Qwen3-4B-Base",
         "Qwen/Qwen3-4B-Base",
         "unsloth/Qwen3-4B-Base-bnb-4bit",
     ),
-    "unsloth/Qwen3-8B-Base-unsloth-bnb-4bit" : (
+    "unsloth/Qwen3-8B-Base-unsloth-bnb-4bit": (
         "unsloth/Qwen3-8B-Base",
         "Qwen/Qwen3-8B-Base",
         "unsloth/Qwen3-8B-Base-bnb-4bit",
     ),
-    "unsloth/Qwen3-14B-Base-unsloth-bnb-4bit" : (
+    "unsloth/Qwen3-14B-Base-unsloth-bnb-4bit": (
         "unsloth/Qwen3-14B-Base",
         "Qwen/Qwen3-14B-Base",
         "unsloth/Qwen3-14B-Base-bnb-4bit",
     ),
-    "unsloth/Qwen3-30B-A3B-Base-bnb-4bit" : (
+    "unsloth/Qwen3-30B-A3B-Base-bnb-4bit": (
         "unsloth/Qwen3-30B-A3B-Base",
         "Qwen/Qwen3-30B-A3B-Base",
     ),
-    "unsloth/phi-4-reasoning-unsloth-bnb-4bit" : (
+    "unsloth/phi-4-reasoning-unsloth-bnb-4bit": (
         "unsloth/phi-4-reasoning",
         "microsoft/Phi-4-reasoning",
         "unsloth/phi-4-reasoning-bnb-4bit",
     ),
-    "unsloth/phi-4-reasoning-plus-unsloth-bnb-4bit" : (
+    "unsloth/phi-4-reasoning-plus-unsloth-bnb-4bit": (
         "unsloth/phi-4-reasoning-plus",
         "microsoft/Phi-4-reasoning-plus",
         "unsloth/phi-4-reasoning-plus-bnb-4bit",
     ),
-    "unsloth/phi-4-mini-reasoning-unsloth-bnb-4bit" : (
+    "unsloth/phi-4-mini-reasoning-unsloth-bnb-4bit": (
         "unsloth/phi-4-mini-reasoning",
         "microsoft/Phi-4-mini-reasoning",
         "unsloth/phi-4-mini-reasoning-bnb-4bit",
     ),
-    "unsloth/Phi-4-mini-instruct-unsloth-bnb-4bit" : (
+    "unsloth/Phi-4-mini-instruct-unsloth-bnb-4bit": (
         "unsloth/Phi-4-mini-instruct",
         "microsoft/Phi-4-mini-instruct",
         "unsloth/Phi-4-mini-instruct-bnb-4bit",
     ),
-    "unsloth/orpheus-3b-0.1-pretrained-unsloth-bnb-4bit" : (
+    "unsloth/orpheus-3b-0.1-pretrained-unsloth-bnb-4bit": (
         "unsloth/orpheus-3b-0.1-pretrained",
         "canopylabs/orpheus-3b-0.1-pretrained",
         "unsloth/orpheus-3b-0.1-pretrained-bnb-4bit",
     ),
-    "unsloth/orpheus-3b-0.1-ft-unsloth-bnb-4bit" : (
+    "unsloth/orpheus-3b-0.1-ft-unsloth-bnb-4bit": (
         "unsloth/orpheus-3b-0.1-ft",
         "canopylabs/orpheus-3b-0.1-ft",
         "unsloth/orpheus-3b-0.1-ft-bnb-4bit",
     ),
-    "unsloth/csm-1b" : (
+    "unsloth/csm-1b": (
         "unsloth/csm-1b",
         "sesame/csm-1b",
     ),
-    "unsloth/whisper-large-v3" : (
+    "unsloth/whisper-large-v3": (
         "unsloth/whisper-large-v3",
         "openai/whisper-large-v3",
     ),
-    "unsloth/whisper-large-v3-turbo" : (
+    "unsloth/whisper-large-v3-turbo": (
         "unsloth/whisper-large-v3-turbo",
         "openai/whisper-large-v3-turbo",
     ),
-    "unsloth/whisper-small" : (
+    "unsloth/whisper-small": (
         "unsloth/whisper-small",
         "openai/whisper-small",
     ),
-    "unsloth/CrisperWhisper" : (
+    "unsloth/CrisperWhisper": (
         "unsloth/CrisperWhisper",
         "nyrahealth/CrisperWhisper",
     ),
-    "unsloth/Llasa-1B" : (
+    "unsloth/Llasa-1B": (
         "unsloth/Llasa-1B",
         "HKUSTAudio/Llasa-1B",
     ),
-    "unsloth/Spark-TTS-0.5B" : (
+    "unsloth/Spark-TTS-0.5B": (
         "unsloth/Spark-TTS-0.5B",
         "SparkAudio/Spark-TTS-0.5B",
     ),
-    "unsloth/Llama-OuteTTS-1.0-1B" : (
+    "unsloth/Llama-OuteTTS-1.0-1B": (
         "unsloth/Llama-OuteTTS-1.0-1B",
         "OuteAI/Llama-OuteTTS-1.0-1B",
     ),
-    "unsloth/medgemma-4b-it-unsloth-bnb-4bit" : (
+    "unsloth/medgemma-4b-it-unsloth-bnb-4bit": (
         "unsloth/medgemma-4b-it",
         "google/medgemma-4b-it",
         "unsloth/medgemma-4b-it-bnb-4bit",
     ),
-    "unsloth/medgemma-27b-text-it-unsloth-bnb-4bit" : (
+    "unsloth/medgemma-27b-text-it-unsloth-bnb-4bit": (
         "unsloth/medgemma-27b-text-it",
         "google/medgemma-27b-text-it",
         "unsloth/medgemma-27b-text-it-bnb-4bit",
     ),
-    "unsloth/Devstral-Small-2505-unsloth-bnb-4bit" : (
+    "unsloth/Devstral-Small-2505-unsloth-bnb-4bit": (
         "unsloth/Devstral-Small-2505",
         "mistralai/Devstral-Small-2505",
         "unsloth/Devstral-Small-2505-bnb-4bit",
     ),
-    "unsloth/DeepSeek-R1-0528-Qwen3-8B-unsloth-bnb-4bit" : (
+    "unsloth/DeepSeek-R1-0528-Qwen3-8B-unsloth-bnb-4bit": (
         "unsloth/DeepSeek-R1-0528-Qwen3-8B",
         "deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
         "unsloth/DeepSeek-R1-0528-Qwen3-8B-bnb-4bit",
     ),
-    "unsloth/Magistral-Small-2506-unsloth-bnb-4bit" : (
+    "unsloth/Magistral-Small-2506-unsloth-bnb-4bit": (
         "unsloth/Magistral-Small-2506",
         "mistralai/Magistral-Small-2506",
         "unsloth/Magistral-Small-2506-bnb-4bit",
     ),
-    "unsloth/Mistral-Small-3.2-24B-Instruct-2506-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Mistral-Small-3.2-24B-Instruct-2506-unsloth-bnb-4bit": {
+        "8": (
             "mistralai/Mistral-Small-3.2-24B-Instruct-2506",
             "unsloth/Mistral-Small-3.2-24B-Instruct-2506-FP8",
             "unsloth/Mistral-Small-3.2-24B-Instruct-2506-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Mistral-Small-3.2-24B-Instruct-2506",
             "mistralai/Mistral-Small-3.2-24B-Instruct-2506",
             "unsloth/Mistral-Small-3.2-24B-Instruct-2506-bnb-4bit",
         ),
     },
-    "unsloth/gemma-3n-E4B-it-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3n-E4B-it-unsloth-bnb-4bit": (
         "unsloth/gemma-3n-E4B-it",
         "google/gemma-3n-E4B-it",
         "unsloth/gemma-3n-E4B-it-unsloth-bnb-4bit",
     ),
-    "unsloth/gemma-3n-E2B-it-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3n-E2B-it-unsloth-bnb-4bit": (
         "unsloth/gemma-3n-E2B-it",
         "google/gemma-3n-E2B-it",
         "unsloth/gemma-3n-E2B-it-unsloth-bnb-4bit",
     ),
-    "unsloth/gemma-3n-E4B-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3n-E4B-unsloth-bnb-4bit": (
         "unsloth/gemma-3n-E4B",
         "google/gemma-3n-E4B",
         "unsloth/gemma-3n-E4B-unsloth-bnb-4bit",
     ),
-    "unsloth/gemma-3n-E2B-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3n-E2B-unsloth-bnb-4bit": (
         "unsloth/gemma-3n-E2B",
         "google/gemma-3n-E2B",
         "unsloth/gemma-3n-E2B-unsloth-bnb-4bit",
     ),
-    "unsloth/Devstral-Small-2507-unsloth-bnb-4bit" : (
+    "unsloth/Devstral-Small-2507-unsloth-bnb-4bit": (
         "unsloth/Devstral-Small-2507",
         "mistralai/Devstral-Small-2507",
         "unsloth/Devstral-Small-2507-bnb-4bit",
     ),
-    "unsloth/Qwen3-30B-A3B-Thinking-2507" : (
+    "unsloth/Qwen3-30B-A3B-Thinking-2507": (
         "unsloth/Qwen3-30B-A3B-Thinking-2507",
         "Qwen/Qwen3-30B-A3B-Thinking-2507",
     ),
-    "unsloth/Qwen3-30B-A3B-Instruct-2507" : (
+    "unsloth/Qwen3-30B-A3B-Instruct-2507": (
         "unsloth/Qwen3-30B-A3B-Instruct-2507",
         "Qwen/Qwen3-30B-A3B-Instruct-2507",
     ),
-    "unsloth/Qwen3-Coder-30B-A3B-Instruct" : (
+    "unsloth/Qwen3-Coder-30B-A3B-Instruct": (
         "unsloth/Qwen3-Coder-30B-A3B-Instruct",
         "Qwen/Qwen3-Coder-30B-A3B-Instruct",
     ),
-    "unsloth/gpt-oss-20b-unsloth-bnb-4bit" : (
+    "unsloth/gpt-oss-20b-unsloth-bnb-4bit": (
         "unsloth/gpt-oss-20b",
         "openai/gpt-oss-20b",
         "unsloth/gpt-oss-20b-unsloth-bnb-4bit",
     ),
-    "unsloth/gpt-oss-120b-unsloth-bnb-4bit" : (
+    "unsloth/gpt-oss-120b-unsloth-bnb-4bit": (
         "unsloth/gpt-oss-120b",
         "openai/gpt-oss-120b",
         "unsloth/gpt-oss-120b-unsloth-bnb-4bit",
     ),
-    "unsloth/Qwen3-4B-Instruct-2507-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-4B-Instruct-2507-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-4B-Instruct-2507-FP8",
             "unsloth/Qwen3-4B-Instruct-2507-FP8",
             "unsloth/Qwen3-4B-Instruct-2507-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-4B-Instruct-2507",
             "Qwen/Qwen3-4B-Instruct-2507",
             "unsloth/Qwen3-4B-Instruct-2507-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-4B-Thinking-2507-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-4B-Thinking-2507-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-4B-Thinking-2507-FP8",
             "unsloth/Qwen3-4B-Thinking-2507-FP8",
             "unsloth/Qwen3-4B-Thinking-2507-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-4B-Thinking-2507",
             "Qwen/Qwen3-4B-Thinking-2507",
             "unsloth/Qwen3-4B-Thinking-2507-bnb-4bit",
         ),
     },
-    "unsloth/gemma-3-270m-it-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-270m-it-unsloth-bnb-4bit": (
         "unsloth/gemma-3-270m-it",
         "google/gemma-3-270m-it",
         "unsloth/gemma-3-270m-it-bnb-4bit",
     ),
-    "unsloth/gemma-3-270m-unsloth-bnb-4bit" : (
+    "unsloth/gemma-3-270m-unsloth-bnb-4bit": (
         "unsloth/gemma-3-270m",
         "google/gemma-3-270m",
         "unsloth/gemma-3-270m-bnb-4bit",
     ),
-    "unsloth/Magistral-Small-2507-unsloth-bnb-4bit" : (
+    "unsloth/Magistral-Small-2507-unsloth-bnb-4bit": (
         "unsloth/Magistral-Small-2507",
         "mistralai/Magistral-Small-2507",
         "unsloth/Magistral-Small-2507-bnb-4bit",
     ),
-    "unsloth/Magistral-Small-2509-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Magistral-Small-2509-unsloth-bnb-4bit": {
+        "8": (
             "mistralai/Magistral-Small-2509",
             "unsloth/Magistral-Small-2509-FP8-Dynamic",
             "unsloth/Magistral-Small-2509-FP8-Dynamic",
         ),
-        "16" : (
+        "16": (
             "unsloth/Magistral-Small-2509",
             "mistralai/Magistral-Small-2509",
             "unsloth/Magistral-Small-2509-bnb-4bit",
         ),
     },
-    "unsloth/Apertus-70B-Instruct-2509-unsloth-bnb-4bit" : (
+    "unsloth/Apertus-70B-Instruct-2509-unsloth-bnb-4bit": (
         "unsloth/Apertus-70B-Instruct-2509",
         "swiss-ai/Apertus-70B-2509",
         "unsloth/Apertus-70B-Instruct-2509-unsloth-bnb-4bit",
     ),
-    "unsloth/Apertus-8B-Instruct-2509-unsloth-bnb-4bit" : (
+    "unsloth/Apertus-8B-Instruct-2509-unsloth-bnb-4bit": (
         "unsloth/Apertus-8B-Instruct-2509",
         "swiss-ai/Apertus-8B-2509",
         "unsloth/Apertus-8B-Instruct-2509-unsloth-bnb-4bit",
     ),
-    "unsloth/granite-4.0-micro-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-micro-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-micro",
         "ibm-granite/granite-4.0-micro",
         "unsloth/granite-4.0-micro-bnb-4bit",
     ),
-    "unsloth/granite-4.0-h-micro-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-h-micro-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-h-micro",
         "ibm-granite/granite-4.0-h-micro",
         "unsloth/granite-4.0-h-micro-bnb-4bit",
     ),
-    "unsloth/granite-4.0-micro-base-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-micro-base-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-micro-base",
         "ibm-granite/granite-4.0-micro-base",
         "unsloth/granite-4.0-micro-base-bnb-4bit",
     ),
-    "unsloth/granite-4.0-h-micro-base-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-h-micro-base-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-h-micro-base",
         "ibm-granite/granite-4.0-h-micro-base",
         "unsloth/granite-4.0-h-micro-base-bnb-4bit",
     ),
-    "unsloth/granite-4.0-h-tiny" : (
+    "unsloth/granite-4.0-h-tiny": (
         "unsloth/granite-4.0-h-tiny",
         "ibm-granite/granite-4.0-h-tiny",
     ),
-    "unsloth/granite-4.0-h-small" : (
+    "unsloth/granite-4.0-h-small": (
         "unsloth/granite-4.0-h-small",
         "ibm-granite/granite-4.0-h-small",
     ),
-    "unsloth/granite-4.0-h-tiny-base" : (
+    "unsloth/granite-4.0-h-tiny-base": (
         "unsloth/granite-4.0-h-tiny-base",
         "ibm-granite/granite-4.0-h-tiny-base",
     ),
-    "unsloth/granite-4.0-h-small-base" : (
+    "unsloth/granite-4.0-h-small-base": (
         "unsloth/granite-4.0-h-small-base",
         "ibm-granite/granite-4.0-h-small-base",
     ),
-    "unsloth/Qwen3-VL-4B-Thinking-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-VL-4B-Thinking-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-VL-4B-Thinking-FP8",
             "unsloth/Qwen3-VL-4B-Thinking-FP8",
             "unsloth/Qwen3-VL-4B-Thinking-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-VL-4B-Thinking",
             "Qwen/Qwen3-VL-4B-Thinking",
             "unsloth/Qwen3-VL-4B-Thinking-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-VL-8B-Thinking-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-VL-8B-Thinking-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-VL-8B-Thinking-FP8",
             "unsloth/Qwen3-VL-8B-Thinking-FP8",
             "unsloth/Qwen3-VL-8B-Thinking-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-VL-8B-Thinking",
             "Qwen/Qwen3-VL-8B-Thinking",
             "unsloth/Qwen3-VL-8B-Thinking-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-VL-4B-Instruct-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-VL-4B-Instruct-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-VL-4B-Instruct-FP8",
             "unsloth/Qwen3-VL-4B-Instruct-FP8",
             "unsloth/Qwen3-VL-4B-Instruct-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-VL-4B-Instruct",
             "Qwen/Qwen3-VL-4B-Instruct",
             "unsloth/Qwen3-VL-4B-Instruct-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-VL-8B-Instruct-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-VL-8B-Instruct-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-VL-8B-Instruct-FP8",
             "unsloth/Qwen3-VL-8B-Instruct-FP8",
             "unsloth/Qwen3-VL-8B-Instruct-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-VL-8B-Instruct",
             "Qwen/Qwen3-VL-8B-Instruct",
             "unsloth/Qwen3-VL-8B-Instruct-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-VL-2B-Thinking-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-VL-2B-Thinking-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-VL-2B-Thinking-FP8",
             "unsloth/Qwen3-VL-2B-Thinking-FP8",
             "unsloth/Qwen3-VL-2B-Thinking-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-VL-2B-Thinking",
             "Qwen/Qwen3-VL-2B-Thinking",
             "unsloth/Qwen3-VL-2B-Thinking-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-VL-32B-Thinking-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-VL-32B-Thinking-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-VL-32B-Thinking-FP8",
             "unsloth/Qwen3-VL-32B-Thinking-FP8",
             "unsloth/Qwen3-VL-32B-Thinking-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-VL-32B-Thinking",
             "Qwen/Qwen3-VL-32B-Thinking",
             "unsloth/Qwen3-VL-32B-Thinking-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-VL-2B-Instruct-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-VL-2B-Instruct-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-VL-2B-Instruct-FP8",
             "unsloth/Qwen3-VL-2B-Instruct-FP8",
             "unsloth/Qwen3-VL-2B-Instruct-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-VL-2B-Instruct",
             "Qwen/Qwen3-VL-2B-Instruct",
             "unsloth/Qwen3-VL-2B-Instruct-bnb-4bit",
         ),
     },
-    "unsloth/Qwen3-VL-32B-Instruct-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Qwen3-VL-32B-Instruct-unsloth-bnb-4bit": {
+        "8": (
             "Qwen/Qwen3-VL-32B-Instruct-FP8",
             "unsloth/Qwen3-VL-32B-Instruct-FP8",
             "unsloth/Qwen3-VL-32B-Instruct-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Qwen3-VL-32B-Instruct",
             "Qwen/Qwen3-VL-32B-Instruct",
             "unsloth/Qwen3-VL-32B-Instruct-bnb-4bit",
         ),
     },
-    "unsloth/granite-4.0-350m-base-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-350m-base-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-350m-base",
         "ibm-granite/granite-4.0-350m-base",
         "unsloth/granite-4.0-350m-base-bnb-4bit",
     ),
-    "unsloth/granite-4.0-350m-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-350m-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-350m",
         "ibm-granite/granite-4.0-350m",
         "unsloth/granite-4.0-350m-bnb-4bit",
     ),
-    "unsloth/granite-4.0-h-350m-base-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-h-350m-base-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-h-350m-base",
         "ibm-granite/granite-4.0-h-350m-base",
         "unsloth/granite-4.0-h-350m-base-bnb-4bit",
     ),
-    "unsloth/granite-4.0-h-350m-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-h-350m-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-h-350m",
         "ibm-granite/granite-4.0-h-350m",
         "unsloth/granite-4.0-h-350m-bnb-4bit",
     ),
-    "unsloth/granite-4.0-1b-base-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-1b-base-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-1b-base",
         "ibm-granite/granite-4.0-1b-base",
         "unsloth/granite-4.0-1b-base-bnb-4bit",
     ),
-    "unsloth/granite-4.0-1b-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-1b-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-1b",
         "ibm-granite/granite-4.0-1b",
         "unsloth/granite-4.0-1b-bnb-4bit",
     ),
-    "unsloth/granite-4.0-h-1b-base-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-h-1b-base-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-h-1b-base",
         "ibm-granite/granite-4.0-h-1b-base",
         "unsloth/granite-4.0-h-1b-base-bnb-4bit",
     ),
-    "unsloth/granite-4.0-h-1b-unsloth-bnb-4bit" : (
+    "unsloth/granite-4.0-h-1b-unsloth-bnb-4bit": (
         "unsloth/granite-4.0-h-1b",
         "ibm-granite/granite-4.0-h-1b",
         "unsloth/granite-4.0-h-1b-bnb-4bit",
     ),
-    "unsloth/gpt-oss-safeguard-20b" : (
+    "unsloth/gpt-oss-safeguard-20b": (
         "unsloth/gpt-oss-safeguard-20b",
         "openai/gpt-oss-safeguard-20b",
     ),
-    "unsloth/gpt-oss-safeguard-120b" : (
+    "unsloth/gpt-oss-safeguard-120b": (
         "unsloth/gpt-oss-safeguard-120b",
         "openai/gpt-oss-safeguard-120b",
     ),
-    "unsloth/functiongemma-270m-it-unsloth-bnb-4bit" : (
+    "unsloth/functiongemma-270m-it-unsloth-bnb-4bit": (
         "unsloth/functiongemma-270m-it",
         "google/functiongemma-270m-it",
         "unsloth/functiongemma-270m-it-unsloth-bnb-4bit",
     ),
     # Ministral 3 models
-    "unsloth/Ministral-3-3B-Instruct-2512-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Ministral-3-3B-Instruct-2512-unsloth-bnb-4bit": {
+        "8": (
             "mistralai/Ministral-3-3B-Instruct-2512",
             "unsloth/Ministral-3-3B-Instruct-2512-FP8",
             "unsloth/Ministral-3-3B-Instruct-2512-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Ministral-3-3B-Instruct-2512",
             "mistralai/Ministral-3-3B-Instruct-2512",
             "unsloth/Ministral-3-3B-Instruct-2512-bnb-4bit",
         ),
     },
-    "unsloth/Ministral-3-3B-Base-2512-unsloth-bnb-4bit" : (
+    "unsloth/Ministral-3-3B-Base-2512-unsloth-bnb-4bit": (
         "unsloth/Ministral-3-3B-Base-2512",
         "mistralai/Ministral-3-3B-Base-2512",
         "unsloth/Ministral-3-3B-Base-2512-bnb-4bit",
     ),
-    "unsloth/Ministral-3-3B-Reasoning-2512-unsloth-bnb-4bit" : (
+    "unsloth/Ministral-3-3B-Reasoning-2512-unsloth-bnb-4bit": (
         "unsloth/Ministral-3-3B-Reasoning-2512",
         "mistralai/Ministral-3-3B-Reasoning-2512",
         "unsloth/Ministral-3-3B-Reasoning-2512-bnb-4bit",
     ),
-    "unsloth/Ministral-3-8B-Instruct-2512-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Ministral-3-8B-Instruct-2512-unsloth-bnb-4bit": {
+        "8": (
             "mistralai/Ministral-3-8B-Instruct-2512",
             "unsloth/Ministral-3-8B-Instruct-2512-FP8",
             "unsloth/Ministral-3-8B-Instruct-2512-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Ministral-3-8B-Instruct-2512",
             "mistralai/Ministral-3-8B-Instruct-2512",
             "unsloth/Ministral-3-8B-Instruct-2512-bnb-4bit",
         ),
     },
-    "unsloth/Ministral-3-8B-Base-2512-unsloth-bnb-4bit" : (
+    "unsloth/Ministral-3-8B-Base-2512-unsloth-bnb-4bit": (
         "unsloth/Ministral-3-8B-Base-2512",
         "mistralai/Ministral-3-8B-Base-2512",
         "unsloth/Ministral-3-8B-Base-2512-bnb-4bit",
     ),
-    "unsloth/Ministral-3-8B-Reasoning-2512-unsloth-bnb-4bit" : (
+    "unsloth/Ministral-3-8B-Reasoning-2512-unsloth-bnb-4bit": (
         "unsloth/Ministral-3-8B-Reasoning-2512",
         "mistralai/Ministral-3-8B-Reasoning-2512",
         "unsloth/Ministral-3-8B-Reasoning-2512-bnb-4bit",
     ),
-    "unsloth/Ministral-3-14B-Instruct-2512-unsloth-bnb-4bit" : {
-        "8" : (
+    "unsloth/Ministral-3-14B-Instruct-2512-unsloth-bnb-4bit": {
+        "8": (
             "mistralai/Ministral-3-14B-Instruct-2512",
             "unsloth/Ministral-3-14B-Instruct-2512-FP8",
             "unsloth/Ministral-3-14B-Instruct-2512-FP8",
         ),
-        "16" : (
+        "16": (
             "unsloth/Ministral-3-14B-Instruct-2512",
             "mistralai/Ministral-3-14B-Instruct-2512",
             "unsloth/Ministral-3-14B-Instruct-2512-bnb-4bit",
         ),
     },
-    "unsloth/Ministral-3-14B-Base-2512-unsloth-bnb-4bit" : (
+    "unsloth/Ministral-3-14B-Base-2512-unsloth-bnb-4bit": (
         "unsloth/Ministral-3-14B-Base-2512",
         "mistralai/Ministral-3-14B-Base-2512",
         "unsloth/Ministral-3-14B-Base-2512-bnb-4bit",
     ),
-    "unsloth/Ministral-3-14B-Reasoning-2512-unsloth-bnb-4bit" : (
+    "unsloth/Ministral-3-14B-Reasoning-2512-unsloth-bnb-4bit": (
         "unsloth/Ministral-3-14B-Reasoning-2512",
         "mistralai/Ministral-3-14B-Reasoning-2512",
         "unsloth/Ministral-3-14B-Reasoning-2512-bnb-4bit",
     ),
-    "unsloth/Kimi-K2-Instruct-BF16" : (
-        "unsloth/Kimi-K2-Instruct",
-    ),
+    "unsloth/Kimi-K2-Instruct-BF16": ("unsloth/Kimi-K2-Instruct",),
 }
 
-INT_TO_FLOAT_MAPPER  = {}
-FLOAT_TO_INT_MAPPER  = {}
+INT_TO_FLOAT_MAPPER = {}
+FLOAT_TO_INT_MAPPER = {}
 MAP_TO_UNSLOTH_16bit = {}
 FLOAT_TO_FP8_BLOCK_MAPPER = {}
-FLOAT_TO_FP8_ROW_MAPPER   = {}
+FLOAT_TO_FP8_ROW_MAPPER = {}
 
 
 def _add_with_lower(mapper, key, value):
@@ -1361,6 +1338,7 @@ def _add_lower_only(mapper, key, value):
         return
     mapper[key.lower()] = value
 
+
 for key, values in __INT_TO_FLOAT_MAPPER.items():
     block, row = None, None
     if type(values) is dict:
diff --git a/unsloth/models/mistral.py b/unsloth/models/mistral.py
index 83e9ab9486..fd8dc57e41 100644
--- a/unsloth/models/mistral.py
+++ b/unsloth/models/mistral.py
@@ -14,7 +14,6 @@
 
 from .llama import *
 import os
-from ._utils import __version__
 from unsloth_zoo.utils import _get_dtype
 from unsloth_zoo.hf_utils import dtype_from_config
 from ..utils.packing import (
@@ -48,7 +47,6 @@ try:
 except:
     MistralSdpaAttention = MistralAttention
     MistralFlashAttention2 = MistralAttention
-from unsloth_zoo.utils import Version, _get_dtype
 
 
 def MistralAttention_fast_forward(
diff --git a/unsloth/models/qwen3.py b/unsloth/models/qwen3.py
index b93dddb186..1d008bd1a8 100644
--- a/unsloth/models/qwen3.py
+++ b/unsloth/models/qwen3.py
@@ -13,9 +13,7 @@
 # limitations under the License.
 
 from .llama import *
-import os
-from ._utils import __version__
-from unsloth_zoo.utils import Version, _get_dtype
+from unsloth_zoo.utils import Version
 from ..utils.packing import get_packed_info_from_kwargs
 from ..utils.attention_dispatch import (
     AttentionConfig,
@@ -48,9 +46,6 @@ except:
             f'Try `pip install --upgrade "transformers>=4.50.3"`\n'
             f"to obtain the latest transformers build, then restart this session."
         )
-from transformers.modeling_attn_mask_utils import (
-    _prepare_4d_causal_attention_mask_for_sdpa,
-)
 
 # For Pytorch 2.1.1
 try:
diff --git a/unsloth/models/qwen3_moe.py b/unsloth/models/qwen3_moe.py
index e1f8c71b6b..efd58fc72e 100644
--- a/unsloth/models/qwen3_moe.py
+++ b/unsloth/models/qwen3_moe.py
@@ -13,8 +13,6 @@
 # limitations under the License.
 
 from .llama import *
-import os
-from ._utils import __version__
 from .llama import (
     LlamaRotaryEmbedding,
     LlamaLinearScalingRotaryEmbedding,
@@ -43,7 +41,6 @@ from transformers.models.qwen3_moe.modeling_qwen3_moe import (
 #     Qwen3SdpaAttention   = Qwen3Attention
 #     Qwen3FlashAttention2 = Qwen3Attention
 # pass
-from unsloth_zoo.utils import Version, _get_dtype
 
 
 torch_nn_functional_softmax = torch.nn.functional.softmax
diff --git a/unsloth/models/rl.py b/unsloth/models/rl.py
index 5651a7da41..c5b167c50c 100755
--- a/unsloth/models/rl.py
+++ b/unsloth/models/rl.py
@@ -18,7 +18,6 @@ __all__ = [
 ]
 
 import torch
-from typing import Any, Callable, Dict, List, Literal, Optional, Tuple, Union
 import inspect
 import os
 import re
@@ -539,8 +538,6 @@ def _wrap_grpo_generate_and_score(trainer_cls):
 
 def _patch_trl_rl_trainers(trainer_file = "grpo_trainer"):
     # Patch for vLLM and Unsloth PEFT
-    import trl
-    import trl.trainer
 
     try:
         trainer = eval(f"trl.trainer.{trainer_file}")
diff --git a/unsloth/models/rl_replacements.py b/unsloth/models/rl_replacements.py
index 9f555416d4..7def25cf94 100755
--- a/unsloth/models/rl_replacements.py
+++ b/unsloth/models/rl_replacements.py
@@ -38,12 +38,7 @@ from unsloth_zoo.log import logger
 from unsloth_zoo.device_type import device_synchronize
 import importlib.util
 from ..device_type import (
-    is_hip,
-    get_device_type,
     DEVICE_TYPE,
-    DEVICE_TYPE_TORCH,
-    DEVICE_COUNT,
-    ALLOW_PREQUANTIZED_MODELS,
 )
 import textwrap
 from ._utils import _get_inference_mode_context_manager
diff --git a/unsloth/models/sentence_transformer.py b/unsloth/models/sentence_transformer.py
index ad59165a50..45f49039ff 100644
--- a/unsloth/models/sentence_transformer.py
+++ b/unsloth/models/sentence_transformer.py
@@ -384,7 +384,7 @@ def _push_to_hub_gguf(
 
     # Save to temporary directory first
     with tempfile.TemporaryDirectory(prefix = "unsloth_st_gguf_") as temp_dir:
-        print(f"Unsloth: Converting SentenceTransformer to GGUF format...")
+        print("Unsloth: Converting SentenceTransformer to GGUF format...")
 
         # Call save_pretrained_gguf to do the local conversion
         result = _save_pretrained_gguf(
@@ -577,9 +577,9 @@ class FastSentenceTransformer(FastModel):
                                 print(f"Pooling mode detected as {mode}, updating...")
                             return mode
 
-        except Exception as e:
+        except Exception:
             print(
-                f"Failed to detect pooling mode, not a sentence-transformers model. Using default pooling mode 'mean', this may or may not work."
+                "Failed to detect pooling mode, not a sentence-transformers model. Using default pooling mode 'mean', this may or may not work."
             )
             return "mean"
 
diff --git a/unsloth/models/vision.py b/unsloth/models/vision.py
index f558aa3f00..be2214eb19 100644
--- a/unsloth/models/vision.py
+++ b/unsloth/models/vision.py
@@ -17,7 +17,6 @@ from transformers import (
     BitsAndBytesConfig,
     AutoProcessor,
     AutoTokenizer,
-    AutoModelForCausalLM,
 )
 
 try:
@@ -62,29 +61,22 @@ from unsloth_zoo.patching_utils import patch_model_and_tokenizer
 from unsloth_zoo.training_utils import prepare_model_for_training
 
 from unsloth_zoo.utils import Version
-from transformers import __version__ as transformers_version
 
 import types
 import functools
 import os
 import gc
-import math
-from typing import Optional, Tuple, List, Union
-import re, inspect, sys
-import contextlib
+import inspect
 
 try:
-    from huggingface_hub.utils import get_token
+    pass
 except:
     # Old HF Hub versions <= 0.0.25
-    from huggingface_hub.utils._token import get_token
+    pass
 from ..device_type import (
-    is_hip,
-    get_device_type,
     DEVICE_TYPE,
     DEVICE_TYPE_TORCH,
     DEVICE_COUNT,
-    ALLOW_PREQUANTIZED_MODELS,
 )
 
 __all__ = [
@@ -108,7 +100,7 @@ PRE_COMPILE_INFERENCE = [
     "gpt_oss",
 ]
 
-from transformers import GenerationConfig, CompileConfig, AutoConfig
+from transformers import CompileConfig, AutoConfig
 
 try:
     from transformers import PreTrainedConfig
@@ -627,7 +619,7 @@ class FastBaseModel:
             default_attn_impl = "eager"
         else:
             default_attn_impl = "flex_attention" if flex_attn_impl else "sdpa"
-        if not ("attn_implementation" in kwargs):
+        if "attn_implementation" not in kwargs:
             kwargs["attn_implementation"] = default_attn_impl
         if not supports_sdpa and kwargs.get("attn_implementation") == "sdpa":
             if os.environ.get("UNSLOTH_ENABLE_FLEX_ATTENTION", "0") == "0":
@@ -682,13 +674,13 @@ class FastBaseModel:
             if dtype == torch.bfloat16:
                 if float32_mixed_precision != True:
                     print(
-                        f"Unsloth: Using bfloat16 full finetuning which cuts memory usage by 50%.\n"
-                        f"To enable float32 training, use `float32_mixed_precision = True` during FastLanguageModel.from_pretrained"
+                        "Unsloth: Using bfloat16 full finetuning which cuts memory usage by 50%.\n"
+                        "To enable float32 training, use `float32_mixed_precision = True` during FastLanguageModel.from_pretrained"
                     )
                 else:
                     print(
-                        f"Unsloth: Using full float32 full finetuning. "
-                        f"To enable bfloat16 training to reduce VRAM usage by 50% albeit with a slightly higher loss, do:\n"
+                        "Unsloth: Using full float32 full finetuning. "
+                        "To enable bfloat16 training to reduce VRAM usage by 50% albeit with a slightly higher loss, do:\n"
                         "use `float32_mixed_precision = False` during FastLanguageModel.from_pretrained"
                     )
                     os.environ["UNSLOTH_BFLOAT16_MIXED_PRECISION"] = "1"
diff --git a/unsloth/save.py b/unsloth/save.py
index 1759d86fb1..47a675513b 100644
--- a/unsloth/save.py
+++ b/unsloth/save.py
@@ -38,19 +38,16 @@ from peft.tuners.lora import Linear4bit as Peft_Linear4bit
 from peft.tuners.lora import Linear as Peft_Linear
 from typing import Optional, Callable, Union, List
 import sys
-import requests
 import torch
 import os
 import shutil
 import pickle
 import gc
 from transformers.models.llama.modeling_llama import logger
-from .kernels import fast_dequantize, QUANT_STATE, get_lora_parameters_bias
+from .kernels import fast_dequantize, get_lora_parameters_bias
 import subprocess
 import psutil
 import re
-from transformers.models.llama.modeling_llama import logger
-from .tokenizer_utils import fix_sentencepiece_gguf
 from .models.loader_utils import get_model_name
 from .models._utils import _convert_torchao_model
 from .ollama_template_mappers import OLLAMA_TEMPLATES, MODEL_TO_OLLAMA_TEMPLATE_MAPPER
@@ -1239,7 +1236,7 @@ def save_to_gguf(
         print(
             f"Unsloth: [1] Converting model into {first_conversion_dtype} GGUF format."
         )
-        print(f"This might take 3 minutes...")
+        print("This might take 3 minutes...")
 
         initial_files, is_vlm_update = convert_to_gguf(
             model_name = model_name,
@@ -1362,7 +1359,7 @@ def save_to_gguf(
     else:
         want_full_precision = first_conversion in frozenset(quantization_method)
 
-    print(f"Unsloth: All GGUF conversions completed successfully!")
+    print("Unsloth: All GGUF conversions completed successfully!")
     print(f"Generated files: {all_saved_locations}")
 
     return all_saved_locations, want_full_precision, is_vlm
@@ -2231,7 +2228,7 @@ def unsloth_push_to_hub_gguf(
         cleanup_temp = False
 
     # Step 2: Call save_pretrained_gguf to do the conversion
-    print(f"Unsloth: Converting model to GGUF format...")
+    print("Unsloth: Converting model to GGUF format...")
 
     try:
         # Call save_pretrained_gguf - it returns all the info we need
@@ -2640,13 +2637,11 @@ def unsloth_convert_lora_to_ggml_and_save_locally(
     )
 
 
-from .models.loader_utils import get_model_name
 from unsloth_zoo.saving_utils import (
     merge_and_overwrite_lora,
     prepare_saving,
 )
 from unsloth_zoo.llama_cpp import (
-    install_llama_cpp,
     convert_to_gguf as _convert_to_gguf,
 )
 
@@ -2974,7 +2969,6 @@ def _unsloth_save_torchao_with_given_config(
         AutoModelForImageTextToText,
         AutoProcessor,
     )
-    from torchao import quantize_
 
     if isinstance(torchao_config, TorchAoConfig):
         quantization_config = torchao_config
@@ -3112,7 +3106,6 @@ def not_implemented_save(*args, **kwargs):
 def patch_saving_functions(model, vision = False):
     import inspect
     import types
-    from typing import Callable, Optional, Union, List
 
     # And now re add our saving methods!
     if model.push_to_hub.__name__ == "unsloth_push_to_hub":
diff --git a/unsloth/tokenizer_utils.py b/unsloth/tokenizer_utils.py
index 8be6bb5a5a..09ab936ed2 100644
--- a/unsloth/tokenizer_utils.py
+++ b/unsloth/tokenizer_utils.py
@@ -18,22 +18,11 @@ from transformers import PreTrainedTokenizerFast
 import re
 import os
 from transformers.models.llama.modeling_llama import logger
-from peft import PeftModelForCausalLM
 import torch
-import itertools
-import collections
 import numpy as np
-import gc
-import subprocess
-import psutil
 
 from unsloth_zoo.tokenizer_utils import (
-    mean_of_trained_tokens,
     add_new_tokens,
-    fix_untrained_tokens,
-)
-from unsloth_zoo.training_utils import (
-    fix_zero_training_loss,
 )
 
 __all__ = [
@@ -357,7 +346,7 @@ def fix_sentencepiece_tokenizer(
         from transformers.convert_slow_tokenizer import import_protobuf
 
         sentencepiece_model_pb2 = import_protobuf()
-    except Exception as e:
+    except Exception:
         try:
             import google.protobuf
             from unsloth_zoo.utils import Version
@@ -946,7 +935,6 @@ def get_tokenizer_info(tokenizer) -> dict:
     }
 
 
-import inspect
 from inspect import getsource
 import trl
 import trl.trainer.sft_trainer
@@ -991,7 +979,7 @@ def patch_sft_trainer_tokenizer():
     Patches the trainer with changes
     """
     try:
-        sft_trainer = eval(f"trl.trainer.sft_trainer.SFTTrainer")
+        sft_trainer = eval("trl.trainer.sft_trainer.SFTTrainer")
     except:
         return
     all_imports = dir(trl.trainer.sft_trainer)
diff --git a/unsloth/trainer.py b/unsloth/trainer.py
index eea985e958..b9b6d286e2 100644
--- a/unsloth/trainer.py
+++ b/unsloth/trainer.py
@@ -14,16 +14,13 @@
 
 import logging
 import os
-import psutil
-import warnings
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from typing import Optional, List
 from functools import wraps
 
 import trl
 import inspect
 from trl import SFTTrainer
-from . import is_bfloat16_supported
 from unsloth.utils import (
     configure_padding_free,
     configure_sample_packing,