recover working install files

Allow users to set UNSLOTH_STUDIO_HOME to relocate all studio data
(venvs, assets, outputs, exports, auth, cache, tensorboard runs).
Defaults to ~/.unsloth/studio when unset (no behaviour change).

studio/backend/core/export/worker.py

@@ -49,9 +49,30 @@ def _activate_transformers_version(model_name: str) -> None:
 
     resolved = _resolve_base_model(model_name)
     if needs_transformers_5(resolved):
-        if not _ensure_venv_t5_exists():
+        from utils.paths.storage_roots import venv_t5_root
-            raise RuntimeError(
+        venv_t5 = str(venv_t5_root())
-                f"Cannot activate transformers 5.x: .venv_t5 missing at {_VENV_T5_DIR}"
+        if os.path.isdir(venv_t5):
+            sys.path.insert(0, venv_t5)
+            logger.info("Activated transformers 5.x from %s", venv_t5)
+        else:
+            # Fallback: pip install at runtime (slower, ~10-15s)
+            logger.warning(".venv_t5 not found at %s — installing at runtime", venv_t5)
+            import subprocess as sp
+
+            os.makedirs(venv_t5, exist_ok = True)
+            r1 = sp.run(
+                [
+                    sys.executable,
+                    "-m",
+                    "pip",
+                    "install",
+                    "--target",
+                    venv_t5,
+                    "--no-deps",
+                    "transformers==5.3.0",
+                ],
+                stdout = sp.PIPE,
+                stderr = sp.STDOUT,
             )
         if _VENV_T5_DIR not in sys.path:
             sys.path.insert(0, _VENV_T5_DIR)

studio/backend/core/inference/worker.py

@@ -51,9 +51,30 @@ def _activate_transformers_version(model_name: str) -> None:
 
     resolved = _resolve_base_model(model_name)
     if needs_transformers_5(resolved):
-        if not _ensure_venv_t5_exists():
+        from utils.paths.storage_roots import venv_t5_root
-            raise RuntimeError(
+        venv_t5 = str(venv_t5_root())
-                f"Cannot activate transformers 5.x: .venv_t5 missing at {_VENV_T5_DIR}"
+        if os.path.isdir(venv_t5):
+            sys.path.insert(0, venv_t5)
+            logger.info("Activated transformers 5.x from %s", venv_t5)
+        else:
+            # Fallback: pip install at runtime (slower, ~10-15s)
+            logger.warning(".venv_t5 not found at %s — installing at runtime", venv_t5)
+            import subprocess as sp
+
+            os.makedirs(venv_t5, exist_ok = True)
+            r1 = sp.run(
+                [
+                    sys.executable,
+                    "-m",
+                    "pip",
+                    "install",
+                    "--target",
+                    venv_t5,
+                    "--no-deps",
+                    "transformers==5.3.0",
+                ],
+                stdout = sp.PIPE,
+                stderr = sp.STDOUT,
             )
         if _VENV_T5_DIR not in sys.path:
             sys.path.insert(0, _VENV_T5_DIR)

studio/backend/core/training/worker.py

@@ -45,9 +45,30 @@ def _activate_transformers_version(model_name: str) -> None:
 
     resolved = _resolve_base_model(model_name)
     if needs_transformers_5(resolved):
-        if not _ensure_venv_t5_exists():
+        from utils.paths.storage_roots import venv_t5_root
-            raise RuntimeError(
+        venv_t5 = str(venv_t5_root())
-                f"Cannot activate transformers 5.x: .venv_t5 missing at {_VENV_T5_DIR}"
+        if os.path.isdir(venv_t5):
+            sys.path.insert(0, venv_t5)
+            logger.info("Activated transformers 5.x from %s", venv_t5)
+        else:
+            # Fallback: pip install at runtime (slower, ~10-15s)
+            logger.warning(".venv_t5 not found at %s — installing at runtime", venv_t5)
+            import subprocess as sp
+
+            os.makedirs(venv_t5, exist_ok = True)
+            r1 = sp.run(
+                [
+                    sys.executable,
+                    "-m",
+                    "pip",
+                    "install",
+                    "--target",
+                    venv_t5,
+                    "--no-deps",
+                    "transformers==5.3.0",
+                ],
+                stdout = sp.PIPE,
+                stderr = sp.STDOUT,
             )
         if _VENV_T5_DIR not in sys.path:
             sys.path.insert(0, _VENV_T5_DIR)

studio/backend/utils/models/model_config.py

@@ -427,7 +427,8 @@ _VLM_MODEL_TYPES = {
 }
 
 # Pre-computed .venv_t5 path and backend dir for subprocess version switching.
-_VENV_T5_DIR = str(Path.home() / ".unsloth" / "studio" / ".venv_t5")
+from utils.paths.storage_roots import venv_t5_root
+_VENV_T5_DIR = str(venv_t5_root())
 _BACKEND_DIR = str(Path(__file__).resolve().parent.parent.parent)
 
 # Inline script executed in a subprocess with transformers 5.x activated.

studio/backend/utils/paths/__init__.py

@@ -8,6 +8,7 @@ Path utilities for model and dataset handling
 from .path_utils import normalize_path, is_local_path, is_model_cached, get_cache_path
 from .storage_roots import (
     studio_root,
+    venv_t5_root,
     assets_root,
     datasets_root,
     dataset_uploads_root,
@@ -36,6 +37,7 @@ __all__ = [
     "is_model_cached",
     "get_cache_path",
     "studio_root",
+    "venv_t5_root",
     "assets_root",
     "datasets_root",
     "dataset_uploads_root",

studio/backend/utils/paths/storage_roots.py

@@ -9,12 +9,26 @@ import tempfile
 
 
 def studio_root() -> Path:
+    """Studio root: env var > config file > default."""
+    custom = os.environ.get("UNSLOTH_STUDIO_HOME")
+    if custom:
+        return Path(custom).expanduser().resolve()
+    conf = Path.home() / ".unsloth" / "studio_home"
+    if conf.is_file():
+        saved = conf.read_text().strip()
+        if saved:
+            return Path(saved).expanduser().resolve()
     return Path.home() / ".unsloth" / "studio"
 
 
+def venv_t5_root() -> Path:
+    """Pre-installed transformers 5.x directory, respects UNSLOTH_STUDIO_HOME."""
+    return studio_root() / ".venv_t5"
+
+
 def cache_root() -> Path:
     """Central cache directory for all studio downloads (models, datasets, etc.)."""
-    return Path.home() / ".unsloth" / "studio" / "cache"
+    return studio_root() / "cache"
 
 
 def assets_root() -> Path:

studio/backend/utils/transformers_version.py

@@ -62,7 +62,8 @@ TRANSFORMERS_5_VERSION = "5.3.0"
 TRANSFORMERS_DEFAULT_VERSION = "4.57.6"
 
 # Pre-installed directory for transformers 5.x — created by setup.sh / setup.ps1
-_VENV_T5_DIR = str(Path.home() / ".unsloth" / "studio" / ".venv_t5")
+from utils.paths.storage_roots import venv_t5_root
+_VENV_T5_DIR = str(venv_t5_root())
 
 
 def _resolve_base_model(model_name: str) -> str:

studio/install_python_stack.py

@@ -140,15 +140,15 @@ def _bootstrap_uv() -> bool:
     global UV_NEEDS_SYSTEM
     if not shutil.which("uv"):
         return False
-    # Probe: try a dry-run install targeting the current Python explicitly.
+    # Probe: try a dry-run install without --system.
-    # Without --python, uv can ignore the activated venv on some platforms.
+    # If uv can't find a venv it exits with code 2.
     probe = subprocess.run(
-        ["uv", "pip", "install", "--dry-run", "--python", sys.executable, "pip"],
+        ["uv", "pip", "install", "--dry-run", "pip"],
         stdout = subprocess.PIPE,
         stderr = subprocess.STDOUT,
     )
     if probe.returncode != 0:
-        # Retry with --system (some envs need it when uv can't find a venv)
+        # Retry with --system to confirm it works
         probe_sys = subprocess.run(
             ["uv", "pip", "install", "--dry-run", "--system", "pip"],
             stdout = subprocess.PIPE,
@@ -204,10 +204,6 @@ def _build_uv_cmd(args: tuple[str, ...]) -> list[str]:
     cmd = ["uv", "pip", "install"]
     if UV_NEEDS_SYSTEM:
         cmd.append("--system")
-    # Always pass --python so uv targets the correct environment.
-    # Without this, uv can ignore an activated venv and install into
-    # the system Python (observed on Colab and similar environments).
-    cmd.extend(["--python", sys.executable])
     cmd.extend(_translate_pip_args_for_uv(args))
     cmd.append("--torch-backend=auto")
     return cmd

studio/setup.ps1

@@ -966,9 +966,14 @@ if (-not $PythonCmd) {
 
 Write-Host "[OK] Using $PythonCmd ($(& $PythonCmd --version 2>&1))" -ForegroundColor Green
 
-# Always create a .venv for isolation -- even for pip installs.
+# ── Studio home (configurable via UNSLOTH_STUDIO_HOME) ──
-# Created in the repo root (parent of studio/).
+$StudioHome = if ($env:UNSLOTH_STUDIO_HOME) { $env:UNSLOTH_STUDIO_HOME } else { Join-Path $env:USERPROFILE ".unsloth\studio" }
-$VenvDir = Join-Path $env:USERPROFILE ".unsloth\studio\.venv"
+# Persist for future `unsloth studio` runs (survives shell restarts)
+$UnslothDir = Join-Path $env:USERPROFILE ".unsloth"
+if (-not (Test-Path $UnslothDir)) { New-Item -ItemType Directory -Path $UnslothDir -Force | Out-Null }
+Set-Content -Path (Join-Path $UnslothDir "studio_home") -Value $StudioHome -NoNewline
+
+$VenvDir = Join-Path $StudioHome ".venv"
 if (-not (Test-Path $VenvDir)) {
     Write-Host "   Creating virtual environment at $VenvDir..." -ForegroundColor Cyan
     & $PythonCmd -m venv $VenvDir
@@ -1090,7 +1095,7 @@ $ErrorActionPreference = $prevEAP
 # The training subprocess just prepends .venv_t5/ to sys.path -- instant switch.
 Write-Host ""
 Write-Host "   Pre-installing transformers 5.x for newer model support..." -ForegroundColor Cyan
-$VenvT5Dir = Join-Path $env:USERPROFILE ".unsloth\studio\.venv_t5"
+$VenvT5Dir = Join-Path $StudioHome ".venv_t5"
 if (Test-Path $VenvT5Dir) { Remove-Item -Recurse -Force $VenvT5Dir }
 New-Item -ItemType Directory -Path $VenvT5Dir -Force | Out-Null
 $prevEAP_t5 = $ErrorActionPreference

studio/setup.sh

@@ -41,26 +41,13 @@ if [[ "$keynames" == *$'\nCOLAB_'* ]]; then
 fi
 
 # ── Detect whether frontend needs building ──
-# Skip if dist/ exists AND no tracked input is newer than dist/.
+# Only skip when BOTH conditions are true:
-# Checks top-level config/entry files and src/, public/ recursively.
+#   1. We're inside site-packages (PyPI / pip install, not editable)
-# This handles: PyPI installs (dist/ bundled), repeat runs (no changes),
+#   2. dist/ already exists (pre-built in the wheel)
-# and upgrades/pulls (source newer than dist/ triggers rebuild).
+# Otherwise always (re)build — handles upgrades, editable installs, and
-_NEED_FRONTEND_BUILD=true
+# pip-from-source where dist/ was never built.
-if [ -d "$SCRIPT_DIR/frontend/dist" ]; then
+if [[ "$SCRIPT_DIR" == */site-packages/* ]] && [ -d "$SCRIPT_DIR/frontend/dist" ]; then
-    # Check all top-level files (package.json, bun.lock, vite.config.ts, index.html, etc.)
+    echo "✅ Frontend pre-built (PyPI) — skipping Node/npm check."
-    _changed=$(find "$SCRIPT_DIR/frontend" -maxdepth 1 -type f \
-        -newer "$SCRIPT_DIR/frontend/dist" -print -quit 2>/dev/null)
-    # Check src/ and public/ recursively (|| true guards against set -e when dirs are missing)
-    if [ -z "$_changed" ]; then
-        _changed=$(find "$SCRIPT_DIR/frontend/src" "$SCRIPT_DIR/frontend/public" \
-            -type f -newer "$SCRIPT_DIR/frontend/dist" -print -quit 2>/dev/null) || true
-    fi
-    if [ -z "$_changed" ]; then
-        _NEED_FRONTEND_BUILD=false
-    fi
-fi
-if [ "$_NEED_FRONTEND_BUILD" = false ]; then
-    echo "✅ Frontend already built and up to date -- skipping Node/npm check."
 else
 NEED_NODE=true
 if command -v node &>/dev/null && command -v npm &>/dev/null; then
@@ -159,27 +146,12 @@ run_quiet "npm run build" npm run build
 
 _restore_gitignores
 trap - EXIT
-
+cd "$SCRIPT_DIR/backend/core/data_recipe/oxc-validator"
-# Validate CSS output -- catch truncated Tailwind builds
+run_quiet "npm install (oxc validator runtime)" npm install
-_MAX_CSS=$(find "$SCRIPT_DIR/frontend/dist/assets" -name '*.css' -exec wc -c {} + 2>/dev/null | sort -n | tail -1 | awk '{print $1}')
-if [ -z "$_MAX_CSS" ]; then
-    echo "⚠️  WARNING: No CSS files were emitted. The frontend build may have failed."
-elif [ "$_MAX_CSS" -lt 100000 ]; then
-    echo "⚠️  WARNING: Largest CSS file is only $((_MAX_CSS / 1024))KB (expected >100KB)."
-    echo "   Tailwind may not have scanned all source files. Check for .gitignore interference."
-fi
-
 cd "$SCRIPT_DIR"
 echo "✅ Frontend built to frontend/dist"
 
-fi  # end frontend build check
+fi  # end frontend dist check
-
-# ── oxc-validator runtime (needs npm -- skip if not available) ──
-if [ -d "$SCRIPT_DIR/backend/core/data_recipe/oxc-validator" ] && command -v npm &>/dev/null; then
-    cd "$SCRIPT_DIR/backend/core/data_recipe/oxc-validator"
-    run_quiet "npm install (oxc validator runtime)" npm install
-    cd "$SCRIPT_DIR"
-fi
 
 # ── 6. Python venv + deps ──
 
@@ -251,261 +223,188 @@ install_python_stack() {
     python "$SCRIPT_DIR/install_python_stack.py"
 }
 
-# Create venv under ~/.unsloth/studio/ (shared location, not in repo).
+if [ "$IS_COLAB" = true ]; then
-# All platforms (including Colab) use the same isolated venv so that
+    # Colab: install packages directly without venv
-# studio dependencies are never installed into the system Python.
+    install_python_stack
-STUDIO_HOME="$HOME/.unsloth/studio"
-VENV_DIR="$STUDIO_HOME/.venv"
-VENV_T5_DIR="$STUDIO_HOME/.venv_t5"
-mkdir -p "$STUDIO_HOME"
-
-# Clean up legacy in-repo venvs if they exist
-[ -d "$REPO_ROOT/.venv" ] && rm -rf "$REPO_ROOT/.venv"
-[ -d "$REPO_ROOT/.venv_overlay" ] && rm -rf "$REPO_ROOT/.venv_overlay"
-[ -d "$REPO_ROOT/.venv_t5" ] && rm -rf "$REPO_ROOT/.venv_t5"
-
-rm -rf "$VENV_DIR"
-rm -rf "$VENV_T5_DIR"
-# Try creating venv with pip; fall back to --without-pip + bootstrap
-# (some environments like Colab have broken ensurepip)
-if ! "$BEST_PY" -m venv "$VENV_DIR" 2>/dev/null; then
-    "$BEST_PY" -m venv --without-pip "$VENV_DIR"
-    source "$VENV_DIR/bin/activate"
-    curl -sS https://bootstrap.pypa.io/get-pip.py | python > /dev/null
 else
+    # Local: create venv under studio home (shared location, not in repo)
+    # Configurable via UNSLOTH_STUDIO_HOME; defaults to ~/.unsloth/studio
+    STUDIO_HOME="${UNSLOTH_STUDIO_HOME:-$HOME/.unsloth/studio}"
+    echo "   Studio home: $STUDIO_HOME"
+    # Persist for future `unsloth studio` runs (survives shell restarts)
+    mkdir -p "$HOME/.unsloth"
+    echo "$STUDIO_HOME" > "$HOME/.unsloth/studio_home"
+    VENV_DIR="$STUDIO_HOME/.venv"
+    VENV_T5_DIR="$STUDIO_HOME/.venv_t5"
+    mkdir -p "$STUDIO_HOME"
+
+    # Clean up legacy in-repo venvs if they exist
+    [ -d "$REPO_ROOT/.venv" ] && rm -rf "$REPO_ROOT/.venv"
+    [ -d "$REPO_ROOT/.venv_overlay" ] && rm -rf "$REPO_ROOT/.venv_overlay"
+    [ -d "$REPO_ROOT/.venv_t5" ] && rm -rf "$REPO_ROOT/.venv_t5"
+
+    rm -rf "$VENV_DIR"
+    rm -rf "$VENV_T5_DIR"
+    "$BEST_PY" -m venv "$VENV_DIR"
     source "$VENV_DIR/bin/activate"
-fi
+    cd "$SCRIPT_DIR"
+    install_python_stack
 
-# ── Ensure uv is available (much faster than pip) ──
+    # ── 6b. Pre-install transformers 5.x into .venv_t5/ ──
-USE_UV=false
+    # Models like GLM-4.7-Flash need transformers>=5.3.0. Instead of pip-installing
-if command -v uv &>/dev/null; then
+    # at runtime (slow, ~10-15s), we pre-install into a separate directory.
-    USE_UV=true
+    # The training subprocess just prepends .venv_t5/ to sys.path — instant switch.
-elif curl -LsSf https://astral.sh/uv/install.sh | sh > /dev/null 2>&1; then
-    export PATH="$HOME/.local/bin:$PATH"
-    command -v uv &>/dev/null && USE_UV=true
-fi
-
-# Helper: install a package, preferring uv with pip fallback
-fast_install() {
-    if [ "$USE_UV" = true ]; then
-        uv pip install --python "$(command -v python)" "$@" && return 0
-    fi
-    python -m pip install "$@"
-}
-
-cd "$SCRIPT_DIR"
-install_python_stack
-
-# ── 6b. Pre-install transformers 5.x into .venv_t5/ ──
-# Models like GLM-4.7-Flash need transformers>=5.3.0. Instead of pip-installing
-# at runtime (slow, ~10-15s), we pre-install into a separate directory.
-# The training subprocess just prepends .venv_t5/ to sys.path -- instant switch.
-echo ""
-echo "   Pre-installing transformers 5.x for newer model support..."
-mkdir -p "$VENV_T5_DIR"
-run_quiet "install transformers 5.x" fast_install --target "$VENV_T5_DIR" --no-deps "transformers==5.3.0"
-run_quiet "install huggingface_hub for t5" fast_install --target "$VENV_T5_DIR" --no-deps "huggingface_hub==1.7.1"
-run_quiet "install hf_xet for t5" fast_install --target "$VENV_T5_DIR" --no-deps "hf_xet==1.4.2"
-# tiktoken is needed by Qwen-family tokenizers. Install with deps since
-# regex/requests may be missing on Windows.
-run_quiet "install tiktoken for t5" fast_install --target "$VENV_T5_DIR" "tiktoken"
-echo "✅ Transformers 5.x pre-installed to $VENV_T5_DIR/"
-
-# ── 7. WSL: pre-install GGUF build dependencies ──
-# On WSL, sudo requires a password and can't be entered during GGUF export
-# (runs in a non-interactive subprocess). Install build deps here instead.
-if grep -qi microsoft /proc/version 2>/dev/null; then
     echo ""
-    echo "⚠️  WSL detected -- installing build dependencies for GGUF export..."
+    echo "   Pre-installing transformers 5.x for newer model support..."
-    _GGUF_DEPS="pciutils build-essential cmake curl git libcurl4-openssl-dev"
+    mkdir -p "$VENV_T5_DIR"
+    run_quiet "pip install transformers 5.x" pip install --target "$VENV_T5_DIR" --no-deps "transformers==5.3.0"
+    run_quiet "pip install huggingface_hub for t5" pip install --target "$VENV_T5_DIR" --no-deps "huggingface_hub==1.3.0"
+    echo "✅ Transformers 5.x pre-installed to $VENV_T5_DIR/"
 
-    # Try without sudo first (works when already root)
+    # ── 7. WSL: pre-install GGUF build dependencies ──
-    apt-get update -y >/dev/null 2>&1 || true
+    # On WSL, sudo requires a password and can't be entered during GGUF export
-    apt-get install -y $_GGUF_DEPS >/dev/null 2>&1 || true
+    # (runs in a non-interactive subprocess). Install build deps here instead.
-
+    if grep -qi microsoft /proc/version 2>/dev/null; then
-    # Check which packages are still missing
+        echo ""
-    _STILL_MISSING=""
+        echo "⚠️  WSL detected — installing build dependencies for GGUF export..."
-    for _pkg in $_GGUF_DEPS; do
+        echo "   You may be prompted for your password."
-        case "$_pkg" in
+        sudo apt-get update -y
-            build-essential) command -v gcc >/dev/null 2>&1 || _STILL_MISSING="$_STILL_MISSING $_pkg" ;;
+        sudo apt-get install -y build-essential cmake curl git libcurl4-openssl-dev
-            pciutils) command -v lspci >/dev/null 2>&1 || _STILL_MISSING="$_STILL_MISSING $_pkg" ;;
-            libcurl4-openssl-dev) dpkg -s "$_pkg" >/dev/null 2>&1 || _STILL_MISSING="$_STILL_MISSING $_pkg" ;;
-            *) command -v "$_pkg" >/dev/null 2>&1 || _STILL_MISSING="$_STILL_MISSING $_pkg" ;;
-        esac
-    done
-    _STILL_MISSING=$(echo "$_STILL_MISSING" | sed 's/^ *//')
-
-    if [ -z "$_STILL_MISSING" ]; then
         echo "✅ GGUF build dependencies installed"
-    elif command -v sudo >/dev/null 2>&1; then
-        echo ""
-        echo "   !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!"
-        echo "   WARNING: We require sudo elevated permissions to install:"
-        echo "   $_STILL_MISSING"
-        echo "   If you accept, we'll run sudo now, and it'll prompt your password."
-        echo "   !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!"
-        echo ""
-        printf "   Accept? [Y/n] "
-        if [ -r /dev/tty ]; then
-            read -r REPLY </dev/tty || REPLY="y"
-        else
-            REPLY="y"
-        fi
-        case "$REPLY" in
-            [nN]*)
-                echo ""
-                echo "   Please install these packages first, then re-run Unsloth Studio setup:"
-                echo "   sudo apt-get update -y && sudo apt-get install -y $_STILL_MISSING"
-                _SKIP_GGUF_BUILD=true
-                ;;
-            *)
-                sudo apt-get update -y
-                sudo apt-get install -y $_STILL_MISSING
-                echo "✅ GGUF build dependencies installed"
-                ;;
-        esac
-    else
-        echo "   sudo is not available on this system."
-        echo "   Please install as root, then re-run setup:"
-        echo "   apt-get install -y $_STILL_MISSING"
-        _SKIP_GGUF_BUILD=true
     fi
 fi
 
 # ── 8. Build llama.cpp binaries for GGUF inference + export ──
-# Builds at ~/.unsloth/llama.cpp — a single shared location under the user's
+# Disabled: llama.cpp build is commented out for now.
-# home directory. This is used by both the inference server and the GGUF
+# UNCOMMENT the block below to re-enable.
-# export pipeline (unsloth-zoo).
+#
-#   - llama-server: for GGUF model inference
+# # Builds at ~/.unsloth/llama.cpp — a single shared location under the user's
-#   - llama-quantize: for GGUF export quantization (symlinked to root for check_llama_cpp())
+# # home directory. This is used by both the inference server and the GGUF
-UNSLOTH_HOME="$HOME/.unsloth"
+# # export pipeline (unsloth-zoo).
-mkdir -p "$UNSLOTH_HOME"
+# #   - llama-server: for GGUF model inference
-LLAMA_CPP_DIR="$UNSLOTH_HOME/llama.cpp"
+# #   - llama-quantize: for GGUF export quantization (symlinked to root for check_llama_cpp())
-LLAMA_SERVER_BIN="$LLAMA_CPP_DIR/build/bin/llama-server"
+# UNSLOTH_HOME="$HOME/.unsloth"
-if [ "${_SKIP_GGUF_BUILD:-}" = true ]; then
+# mkdir -p "$UNSLOTH_HOME"
-    echo ""
+# LLAMA_CPP_DIR="$UNSLOTH_HOME/llama.cpp"
-    echo "Skipping llama-server build (missing dependencies)"
+# LLAMA_SERVER_BIN="$LLAMA_CPP_DIR/build/bin/llama-server"
-    echo "   Install the missing packages and re-run setup to enable GGUF inference."
+# rm -rf "$LLAMA_CPP_DIR"
-else
+# {
-rm -rf "$LLAMA_CPP_DIR"
+#     # Check prerequisites
-{
+#     if ! command -v cmake &>/dev/null; then
-    # Check prerequisites
+#         echo ""
-    if ! command -v cmake &>/dev/null; then
+#         echo "⚠️  cmake not found — skipping llama-server build (GGUF inference won't be available)"
-        echo ""
+#         echo "   Install cmake and re-run setup.sh to enable GGUF inference."
-        echo "⚠️  cmake not found — skipping llama-server build (GGUF inference won't be available)"
+#     elif ! command -v git &>/dev/null; then
-        echo "   Install cmake and re-run setup.sh to enable GGUF inference."
+#         echo ""
-    elif ! command -v git &>/dev/null; then
+#         echo "⚠️  git not found — skipping llama-server build (GGUF inference won't be available)"
-        echo ""
+#     else
-        echo "⚠️  git not found — skipping llama-server build (GGUF inference won't be available)"
+#         echo ""
-    else
+#         echo "Building llama-server for GGUF inference..."
-        echo ""
+#
-        echo "Building llama-server for GGUF inference..."
+#         BUILD_OK=true
-
+#         run_quiet "clone llama.cpp" git clone --depth 1 https://github.com/ggml-org/llama.cpp.git "$LLAMA_CPP_DIR" || BUILD_OK=false
-        BUILD_OK=true
+#
-        run_quiet "clone llama.cpp" git clone --depth 1 https://github.com/ggml-org/llama.cpp.git "$LLAMA_CPP_DIR" || BUILD_OK=false
+#         if [ "$BUILD_OK" = true ]; then
-
+#             # Skip tests/examples we don't need (faster build)
-        if [ "$BUILD_OK" = true ]; then
+#             CMAKE_ARGS="-DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_EXAMPLES=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_NATIVE=ON"
-            # Skip tests/examples we don't need (faster build)
+#
-            CMAKE_ARGS="-DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_EXAMPLES=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_NATIVE=ON"
+#             # Use ccache if available (dramatically faster rebuilds)
-
+#             if command -v ccache &>/dev/null; then
-            # Use ccache if available (dramatically faster rebuilds)
+#                 CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_CUDA_COMPILER_LAUNCHER=ccache"
-            if command -v ccache &>/dev/null; then
+#                 echo "   Using ccache for faster compilation"
-                CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_CUDA_COMPILER_LAUNCHER=ccache"
+#             fi
-                echo "   Using ccache for faster compilation"
+#
-            fi
+#             # Detect CUDA: check nvcc on PATH, then common install locations
-
+#             NVCC_PATH=""
-            # Detect CUDA: check nvcc on PATH, then common install locations
+#             if command -v nvcc &>/dev/null; then
-            NVCC_PATH=""
+#                 NVCC_PATH="$(command -v nvcc)"
-            if command -v nvcc &>/dev/null; then
+#             elif [ -x /usr/local/cuda/bin/nvcc ]; then
-                NVCC_PATH="$(command -v nvcc)"
+#                 NVCC_PATH="/usr/local/cuda/bin/nvcc"
-            elif [ -x /usr/local/cuda/bin/nvcc ]; then
+#                 export PATH="/usr/local/cuda/bin:$PATH"
-                NVCC_PATH="/usr/local/cuda/bin/nvcc"
+#             elif ls /usr/local/cuda-*/bin/nvcc &>/dev/null 2>&1; then
-                export PATH="/usr/local/cuda/bin:$PATH"
+#                 # Pick the newest cuda-XX.X directory
-            elif ls /usr/local/cuda-*/bin/nvcc &>/dev/null 2>&1; then
+#                 NVCC_PATH="$(ls -d /usr/local/cuda-*/bin/nvcc 2>/dev/null | sort -V | tail -1)"
-                # Pick the newest cuda-XX.X directory
+#                 export PATH="$(dirname "$NVCC_PATH"):$PATH"
-                NVCC_PATH="$(ls -d /usr/local/cuda-*/bin/nvcc 2>/dev/null | sort -V | tail -1)"
+#             fi
-                export PATH="$(dirname "$NVCC_PATH"):$PATH"
+#
-            fi
+#             if [ -n "$NVCC_PATH" ]; then
-
+#                 echo "   Building with CUDA support (nvcc: $NVCC_PATH)..."
-            if [ -n "$NVCC_PATH" ]; then
+#                 CMAKE_ARGS="$CMAKE_ARGS -DGGML_CUDA=ON"
-                echo "   Building with CUDA support (nvcc: $NVCC_PATH)..."
+#
-                CMAKE_ARGS="$CMAKE_ARGS -DGGML_CUDA=ON"
+#                 # Detect GPU compute capability and limit CUDA architectures
-
+#                 # Without this, cmake builds for ALL default archs (very slow)
-                # Detect GPU compute capability and limit CUDA architectures
+#                 CUDA_ARCHS=""
-                # Without this, cmake builds for ALL default archs (very slow)
+#                 if command -v nvidia-smi &>/dev/null; then
-                CUDA_ARCHS=""
+#                     # Read all GPUs, deduplicate (handles mixed-GPU hosts)
-                if command -v nvidia-smi &>/dev/null; then
+#                     _raw_caps=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader 2>/dev/null || true)
-                    # Read all GPUs, deduplicate (handles mixed-GPU hosts)
+#                     while IFS= read -r _cap; do
-                    _raw_caps=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader 2>/dev/null || true)
+#                         _cap=$(echo "$_cap" | tr -d '[:space:]')
-                    while IFS= read -r _cap; do
+#                         if [[ "$_cap" =~ ^([0-9]+)\.([0-9]+)$ ]]; then
-                        _cap=$(echo "$_cap" | tr -d '[:space:]')
+#                             _arch="${BASH_REMATCH[1]}${BASH_REMATCH[2]}"
-                        if [[ "$_cap" =~ ^([0-9]+)\.([0-9]+)$ ]]; then
+#                             # Append if not already present
-                            _arch="${BASH_REMATCH[1]}${BASH_REMATCH[2]}"
+#                             case ";$CUDA_ARCHS;" in
-                            # Append if not already present
+#                                 *";$_arch;"*) ;;
-                            case ";$CUDA_ARCHS;" in
+#                                 *) CUDA_ARCHS="${CUDA_ARCHS:+$CUDA_ARCHS;}$_arch" ;;
-                                *";$_arch;"*) ;;
+#                             esac
-                                *) CUDA_ARCHS="${CUDA_ARCHS:+$CUDA_ARCHS;}$_arch" ;;
+#                         fi
-                            esac
+#                     done <<< "$_raw_caps"
-                        fi
+#                 fi
-                    done <<< "$_raw_caps"
+#
-                fi
+#                 if [ -n "$CUDA_ARCHS" ]; then
-
+#                     echo "   GPU compute capabilities: ${CUDA_ARCHS//;/, } -- limiting build to detected archs"
-                if [ -n "$CUDA_ARCHS" ]; then
+#                     CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_ARCHITECTURES=${CUDA_ARCHS}"
-                    echo "   GPU compute capabilities: ${CUDA_ARCHS//;/, } -- limiting build to detected archs"
+#                 else
-                    CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_ARCHITECTURES=${CUDA_ARCHS}"
+#                     echo "   Could not detect GPU arch -- building for all default CUDA architectures (slower)"
-                else
+#                 fi
-                    echo "   Could not detect GPU arch -- building for all default CUDA architectures (slower)"
+#
-                fi
+#                 # Multi-threaded nvcc compilation (uses all CPU cores per .cu file)
-
+#                 CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_FLAGS=--threads=0"
-                # Multi-threaded nvcc compilation (uses all CPU cores per .cu file)
+#             elif [ -d /usr/local/cuda ] || nvidia-smi &>/dev/null; then
-                CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CUDA_FLAGS=--threads=0"
+#                 echo "   CUDA driver detected but nvcc not found — building CPU-only"
-            elif [ -d /usr/local/cuda ] || nvidia-smi &>/dev/null; then
+#                 echo "   To enable GPU: install cuda-toolkit or add nvcc to PATH"
-                echo "   CUDA driver detected but nvcc not found — building CPU-only"
+#             else
-                echo "   To enable GPU: install cuda-toolkit or add nvcc to PATH"
+#                 echo "   Building CPU-only (no CUDA detected)..."
-            else
+#             fi
-                echo "   Building CPU-only (no CUDA detected)..."
+#
-            fi
+#             NCPU=$(nproc 2>/dev/null || sysctl -n hw.ncpu 2>/dev/null || echo 4)
-
+#
-            NCPU=$(nproc 2>/dev/null || sysctl -n hw.ncpu 2>/dev/null || echo 4)
+#             # Use Ninja if available (faster parallel builds than Make)
-
+#             CMAKE_GENERATOR_ARGS=""
-            # Use Ninja if available (faster parallel builds than Make)
+#             if command -v ninja &>/dev/null; then
-            CMAKE_GENERATOR_ARGS=""
+#                 CMAKE_GENERATOR_ARGS="-G Ninja"
-            if command -v ninja &>/dev/null; then
+#             fi
-                CMAKE_GENERATOR_ARGS="-G Ninja"
+#
-            fi
+#             run_quiet "cmake llama.cpp" cmake $CMAKE_GENERATOR_ARGS -S "$LLAMA_CPP_DIR" -B "$LLAMA_CPP_DIR/build" $CMAKE_ARGS || BUILD_OK=false
-
+#         fi
-            run_quiet "cmake llama.cpp" cmake $CMAKE_GENERATOR_ARGS -S "$LLAMA_CPP_DIR" -B "$LLAMA_CPP_DIR/build" $CMAKE_ARGS || BUILD_OK=false
+#
-        fi
+#         if [ "$BUILD_OK" = true ]; then
-
+#             run_quiet "build llama-server" cmake --build "$LLAMA_CPP_DIR/build" --config Release --target llama-server -j"$NCPU" || BUILD_OK=false
-        if [ "$BUILD_OK" = true ]; then
+#         fi
-            run_quiet "build llama-server" cmake --build "$LLAMA_CPP_DIR/build" --config Release --target llama-server -j"$NCPU" || BUILD_OK=false
+#
-        fi
+#         # Also build llama-quantize (needed by unsloth-zoo's GGUF export pipeline)
-
+#         if [ "$BUILD_OK" = true ]; then
-        # Also build llama-quantize (needed by unsloth-zoo's GGUF export pipeline)
+#             run_quiet "build llama-quantize" cmake --build "$LLAMA_CPP_DIR/build" --config Release --target llama-quantize -j"$NCPU" || true
-        if [ "$BUILD_OK" = true ]; then
+#             # Symlink to llama.cpp root — check_llama_cpp() looks for the binary there
-            run_quiet "build llama-quantize" cmake --build "$LLAMA_CPP_DIR/build" --config Release --target llama-quantize -j"$NCPU" || true
+#             QUANTIZE_BIN="$LLAMA_CPP_DIR/build/bin/llama-quantize"
-            # Symlink to llama.cpp root — check_llama_cpp() looks for the binary there
+#             if [ -f "$QUANTIZE_BIN" ]; then
-            QUANTIZE_BIN="$LLAMA_CPP_DIR/build/bin/llama-quantize"
+#                 ln -sf build/bin/llama-quantize "$LLAMA_CPP_DIR/llama-quantize"
-            if [ -f "$QUANTIZE_BIN" ]; then
+#             fi
-                ln -sf build/bin/llama-quantize "$LLAMA_CPP_DIR/llama-quantize"
+#         fi
-            fi
+#
-        fi
+#         if [ "$BUILD_OK" = true ]; then
-
+#             if [ -f "$LLAMA_SERVER_BIN" ]; then
-        if [ "$BUILD_OK" = true ]; then
+#                 echo "✅ llama-server built at $LLAMA_SERVER_BIN"
-            if [ -f "$LLAMA_SERVER_BIN" ]; then
+#             else
-                echo "✅ llama-server built at $LLAMA_SERVER_BIN"
+#                 echo "⚠️  llama-server binary not found after build — GGUF inference won't be available"
-            else
+#             fi
-                echo "⚠️  llama-server binary not found after build — GGUF inference won't be available"
+#             if [ -f "$LLAMA_CPP_DIR/llama-quantize" ]; then
-            fi
+#                 echo "✅ llama-quantize available for GGUF export"
-            if [ -f "$LLAMA_CPP_DIR/llama-quantize" ]; then
+#             fi
-                echo "✅ llama-quantize available for GGUF export"
+#         else
-            fi
+#             echo "⚠️  llama-server build failed — GGUF inference won't be available, but everything else works"
-        else
+#         fi
-            echo "⚠️  llama-server build failed — GGUF inference won't be available, but everything else works"
+#     fi
-        fi
+# }
-    fi
-}
-fi  # end _SKIP_GGUF_BUILD check
 
 echo ""
 if [ "$IS_COLAB" = true ]; then
@@ -514,9 +413,6 @@ if [ "$IS_COLAB" = true ]; then
     echo "╠══════════════════════════════════════╣"
     echo "║ Unsloth Studio is ready to start     ║"
     echo "║ in your Colab notebook!              ║"
-    echo "║                                      ║"
-    echo "║ from colab import start              ║"
-    echo "║ start()                              ║"
     echo "╚══════════════════════════════════════╝"
 else
     echo "╔══════════════════════════════════════╗"
@@ -524,6 +420,6 @@ else
     echo "╠══════════════════════════════════════╣"
     echo "║ Launch with:                         ║"
     echo "║                                      ║"
-    echo "║ unsloth studio -H 0.0.0.0 -p 8888    ║"
+    echo "║ unsloth studio -H 0.0.0.0 -p 8000    ║"
     echo "╚══════════════════════════════════════╝"
 fi

unsloth_cli/commands/studio.py

@@ -12,7 +12,18 @@ import typer
 
 studio_app = typer.Typer(help = "Unsloth Studio commands.")
 
-STUDIO_HOME = Path.home() / ".unsloth" / "studio"
+
+def _studio_home() -> Path:
+    """Studio root: env var > config file > default."""
+    custom = os.environ.get("UNSLOTH_STUDIO_HOME")
+    if custom:
+        return Path(custom).expanduser().resolve()
+    conf = Path.home() / ".unsloth" / "studio_home"
+    if conf.is_file():
+        saved = conf.read_text().strip()
+        if saved:
+            return Path(saved).expanduser().resolve()
+    return Path.home() / ".unsloth" / "studio"
 
 # __file__ is unsloth_cli/commands/studio.py -- two parents up is the package root
 # (either site-packages or the repo root for editable installs).
@@ -22,9 +33,9 @@ _PACKAGE_ROOT = Path(__file__).resolve().parent.parent.parent
 def _studio_venv_python() -> Optional[Path]:
     """Return the studio venv Python binary, or None if not set up."""
     if platform.system() == "Windows":
-        p = STUDIO_HOME / ".venv" / "Scripts" / "python.exe"
+        p = _studio_home() / ".venv" / "Scripts" / "python.exe"
     else:
-        p = STUDIO_HOME / ".venv" / "bin" / "python"
+        p = _studio_home() / ".venv" / "bin" / "python"
     return p if p.is_file() else None
 
 
@@ -44,7 +55,7 @@ def _find_run_py() -> Optional[Path]:
         "lib/python*/site-packages/studio/backend/run.py",
         "Lib/site-packages/studio/backend/run.py",
     ):
-        for match in (STUDIO_HOME / ".venv").glob(pattern):
+        for match in (_studio_home() / ".venv").glob(pattern):
             return match
     return None
 
@@ -64,7 +75,7 @@ def _find_setup_script() -> Optional[Path]:
         f"lib/python*/site-packages/studio/{name}",
         f"Lib/site-packages/studio/{name}",
     ):
-        for match in (STUDIO_HOME / ".venv").glob(pattern):
+        for match in (_studio_home() / ".venv").glob(pattern):
             return match
     return None
 
@@ -85,7 +96,7 @@ def studio_default(
         return
 
     # Always use the studio venv if it exists and we're not already in it
-    studio_venv_dir = STUDIO_HOME / ".venv"
+    studio_venv_dir = _studio_home() / ".venv"
     in_studio_venv = sys.prefix.startswith(str(studio_venv_dir))
 
     if not in_studio_venv: