Triton dep in conda installation

[SergioRubio01]

Hey there,
So I have basically followed all the steps mentioned in the README.md for conda installation, but the same issue keeps stopping the installation. I have not tried to install it manually yet from the GitHub repo but I am not quite sure if this should be done eitherway.

This is the code and issue:

conda create --name unsloth_env \
    python=3.11 \
    pytorch-cuda=12.1 \
    pytorch cudatoolkit xformers -c pytorch -c nvidia -c xformers \
    -y
conda activate unsloth_env

pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
pip install --no-deps trl peft accelerate bitsandbytes

And the issue:

  Using cached tokenizers-0.20.3-cp312-none-win_amd64.whl.metadata (6.9 kB)
Collecting safetensors>=0.4.1 (from transformers<=4.46.3,>=4.46.1->unsloth@ git+https://github.com/unslothai/unsloth.git->unsloth[colab-new]@ git+https://github.com/unslothai/unsloth.git)
  Using cached safetensors-0.4.5-cp312-none-win_amd64.whl.metadata (3.9 kB)
INFO: pip is looking at multiple versions of unsloth-zoo to determine which version is compatible with other requirements. This could take a while.
Collecting unsloth_zoo>=2024.11.8 (from unsloth@ git+https://github.com/unslothai/unsloth.git->unsloth[colab-new]@ git+https://github.com/unslothai/unsloth.git)
  Using cached unsloth_zoo-2024.12.1-py3-none-any.whl.metadata (16 kB)
  Using cached unsloth_zoo-2024.11.8-py3-none-any.whl.metadata (16 kB)
ERROR: Cannot install unsloth because these package versions have conflicting dependencies.

The conflict is caused by:
    unsloth-zoo 2024.12.3 depends on triton
    unsloth-zoo 2024.12.1 depends on triton
    unsloth-zoo 2024.11.8 depends on triton

To fix this you could try to:
1. loosen the range of package versions you've specified
2. remove package versions to allow pip to attempt to solve the dependency conflict      

ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/topics/dependency-resolution/#dealing-with-dependency-conflicts

Isn't the fast installation suppossed to do this without deps conflict?

Thanks!

[SergioRubio01]

Found that by installing triton for python 3.11 as:

pip install https://github.com/woct0rdho/triton-windows/releases/download/v3.1.0-windows.post5/triton-3.1.0-cp311-cp311-win_amd64.whl

triton is finally installed in the conda env. If anyone finds a better way please let me know

[danielhanchen]

Apologies on the delay - I'm trying to make Windows a more seamless experience since in the meantime Unsloth only supports Linux.

I edited some pyproject.toml files - for Windows installations, it might be doable as of today, but I'm not 100% sure.