Skip to content

Installation instructions for CUDA #20

Description

@sliterok

I've been trying to get this repo running for hours and I'm hitting a wall of dependency conflicts. Could you clarify the exact verified combination of torch and cuda versions to use?

My setup:

  • GPU: NVIDIA RTX 3090
  • CUDA: 13.2
  • OS: Windows 11

Issues encountered:

  1. AttributeError: module 'torch' has no attribute 'float8_e8m0fnu' when using latest stable torch (2.6.0).
  2. Trying to use torch nightly (2.13.0+) to fix the above leads to RuntimeError: Configured CUDA binary not found at ... libbitsandbytes_cuda132.dll in bitsandbytes.
  3. pyproject.toml defines torch>=2.11, which doesn't exist on standard CUDA channels, causing pip to pull CPU-only versions or conflict with torchaudio/torchvision.

What is the officially supported environment to run this without patching bitsandbytes or hacking pyproject.toml?

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions