diff --git a/README.md b/README.md
index 45312a43d1..4bdd7e2893 100644
--- a/README.md
+++ b/README.md
@@ -193,7 +193,7 @@ print(f'pip install --upgrade pip && pip install "unsloth[{x}] @ git+https://git
 ### Windows Installation
 
 To run Unsloth directly on Windows:
-- Install Triton from this Windows fork and follow the instructions: https://github.com/woct0rdho/triton-windows
+- Install Triton from this Windows fork and follow the instructions: https://github.com/woct0rdho/triton-windows (be aware that the Windows fork requires PyTorch >= 2.4 and CUDA 12)
 - In the SFTTrainer, set `dataset_num_proc=1` to avoid a crashing issue:
 ```python
 trainer = SFTTrainer(
@@ -202,12 +202,15 @@ trainer = SFTTrainer(
 )
 ```
 
+### Advanced/Troubleshooting
+
 For **advanced installation instructions** or if you see weird errors during installations:
 
 1. Install `torch` and `triton`. Go to https://pytorch.org to install it. For example `pip install torch torchvision torchaudio triton`
 2. Confirm if CUDA is installated correctly. Try `nvcc`. If that fails, you need to install `cudatoolkit` or CUDA drivers.
 3. Install `xformers` manually. You can try installing `vllm` and seeing if `vllm` succeeds. Check if `xformers` succeeded with `python -m xformers.info` Go to https://github.com/facebookresearch/xformers. Another option is to install `flash-attn` for Ampere GPUs.
-4.  Finally, install `bitsandbytes` and check it with `python -m bitsandbytes`
+4. Double check that your versions of Python, CUDA, CUDNN, `torch`, `triton`, and `xformers` are compatible with one another. The [PyTorch Compatibility Matrix](https://github.com/pytorch/pytorch/blob/main/RELEASE.md#release-compatibility-matrix) may be useful. 
+5. Finally, install `bitsandbytes` and check it with `python -m bitsandbytes`
 
 ## 📜 [Documentation](https://docs.unsloth.ai)
 - Go to our official [Documentation](https://docs.unsloth.ai) for saving to GGUF, checkpointing, evaluation and more!
diff --git a/pyproject.toml b/pyproject.toml
index 59a7c44737..07085adcca 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -196,6 +196,10 @@ cu126onlytorch260 = [
     "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 platform_system == 'Linux'",
     "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 platform_system == 'Linux'",
     "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 platform_system == 'Linux'",
+    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp39-cp39-win_amd64.whl ; python_version=='3.9' and platform_system == 'Windows'",
+    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp310-cp310-win_amd64.whl ; python_version=='3.10' and platform_system == 'Windows'",
+    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp311-cp311-win_amd64.whl ; python_version=='3.11' and platform_system == 'Windows'",
+    "xformers @ https://download.pytorch.org/whl/cu126/xformers-0.0.29.post3-cp312-cp312-win_amd64.whl ; python_version=='3.12' and platform_system == 'Windows'",
 ]
 cu118 = [
     "unsloth[huggingface]",
diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 27651be97a..909dfc339b 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -775,9 +775,12 @@ def LlamaModel_fast_forward(
             self.SWA_mask = True
             self.GA_mask  = False
         elif attention_mask is not None:
-
             # Fixes https://github.com/unslothai/unsloth/issues/853
             # Unsloth needs a 2D mask, not a [2, 1, n, n] mask!
+
+            # https://github.com/pytorch/pytorch/issues/103749
+            # Need to convert to float and not using bool
+            attention_mask = (1.0 - attention_mask.float()) * torch.finfo(inputs_embeds.dtype).min
             dynamic_SWA_mask = _prepare_4d_causal_attention_mask_for_sdpa(
                 attention_mask,
                 (batch_size, seq_length),
diff --git a/unsloth/models/loader.py b/unsloth/models/loader.py
index 39b367e275..186545cf0c 100644
--- a/unsloth/models/loader.py
+++ b/unsloth/models/loader.py
@@ -24,10 +24,14 @@ from peft import PeftConfig, PeftModel
 from .loader_utils import get_model_name
 import os, contextlib, sys
 try:
-    from huggingface_hub.utils import get_token
+    from huggingface_hub import get_token
 except:
-    # Old HF Hub versions <= 0.0.25
-    from huggingface_hub.utils._token import get_token
+    try:
+        from huggingface_hub.utils import get_token
+    except:
+        # For older versions of huggingface_hub
+        from huggingface_hub.utils._token import get_token
+    pass
 pass
 from huggingface_hub import HfFileSystem
 import importlib.util
diff --git a/unsloth/save.py b/unsloth/save.py
index 0f75ecfd05..eaddfa05c5 100644
--- a/unsloth/save.py
+++ b/unsloth/save.py
@@ -31,10 +31,14 @@ from transformers.models.llama.modeling_llama import logger
 from .tokenizer_utils import fix_sentencepiece_gguf
 from huggingface_hub import HfApi
 try:
-    from huggingface_hub.utils import get_token
+    from huggingface_hub import get_token
 except:
-    # Old HF Hub versions <= 0.0.25
-    from huggingface_hub.utils._token import get_token
+    try:
+        from huggingface_hub.utils import get_token
+    except:
+        # For older versions of huggingface_hub
+        from huggingface_hub.utils._token import get_token
+    pass
 pass
 from pathlib import Path