All pull requests

Pull requests

• Env var base dir

  #25·Tauheed-Elahee opened Mar 13, 2026·

  ·

• Remove NVIDIA CUDA 12.1 hard pins from requirements

  #24·Tauheed-Elahee opened Mar 13, 2026·

  ·

• Default cuda 0

  #23·Tauheed-Elahee opened Mar 13, 2026·

  ·

• Bleeding Edge - Accept All Pull Requests

  #22·Tauheed-Elahee opened Mar 13, 2026·

  ·
  1

• Fix incorrect --clustering_type arg in README examples

  #20·Tauheed-Elahee opened Mar 13, 2026·

  ·

• Add compatibility for newer sae_lens forward() return type

  #19·Tauheed-Elahee opened Mar 13, 2026·

  ·

• Use dynamic device count instead of hardcoded num_devices=2

  #18·Tauheed-Elahee opened Mar 13, 2026·

  ·

• Fix TransformerLens model name for GPT-2

  #17·Tauheed-Elahee opened Mar 13, 2026·

  ·

• Update torch.load calls to include weights_only parameter

  #16·Tauheed-Elahee opened Mar 12, 2026·

  ·

• legendHandles => legend_handles

  #13·JasonGross opened Sep 14, 2024Contributor·

  ·
  1

• Revert to cpu if cuda is not available (simpler version)

  #11·JasonGross opened Sep 9, 2024Contributor·

  ·

• Add support for loading 16-bit models with --dtype

  #10·JasonGross opened Sep 9, 2024Contributor·

  ·

• README: add some hyperlinks and code font

  #9·JasonGross opened Sep 7, 2024Contributor·

  ·

• Fix some typos in README.md

  #8·JasonGross opened Sep 7, 2024Contributor·

  ·

• Revert to cpu if cuda is not available (backwards compatible version)

  #7·JasonGross opened Sep 7, 2024Contributor·

  ·

• Fix to better support MPS: convert to float32 earlier

  #6·JasonGross opened Sep 7, 2024Contributor·

  ·

• Use pathlib more and have default cache paths be relative to the repo directory

  #5·JasonGross opened Sep 7, 2024Contributor·

  ·

• Don't fail with "AssertionError: Not enough CUDA devices to support n_devices 2" when there aren't 2 GPUs

  #4·JasonGross opened Aug 28, 2024Contributor·

  ·