Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix llama model sdpa attention forward function masking bug when output_attentions=True #30652

Merged
merged 14 commits into from
May 15, 2024
Merged

Fix llama model sdpa attention forward function masking bug when output_attentions=True #30652

merged 14 commits into from
May 15, 2024

Commits on May 4, 2024

  1. Fix llama model forward function with attention=True, same-length enc… 

    …oded sequence.
    @Aladoro
    Aladoro committed May 4, 2024
    Configuration menu
    Copy the full SHA
    514c1c3 View commit details
    Browse the repository at this point in the history

  2. Fix style

    @Aladoro
    Aladoro committed May 4, 2024
    Configuration menu
    Copy the full SHA
    6c0fa7b View commit details
    Browse the repository at this point in the history

  3. propagate fix to modeling_cohere, gemma, dbrx, and olmo (which copy t… 

    …he same sdpa masking logic from llama)
    @Aladoro
    Aladoro committed May 4, 2024
    Configuration menu
    Copy the full SHA
    0d91bea View commit details
    Browse the repository at this point in the history

  4. Fix style

    @Aladoro
    Aladoro committed May 4, 2024
    Configuration menu
    Copy the full SHA
    894c14b View commit details
    Browse the repository at this point in the history

Commits on May 6, 2024

  1. ignore unnecessary sdpa mask converter when output_attentions=True

    @Aladoro
    Aladoro committed May 6, 2024
    Configuration menu
    Copy the full SHA
    2584308 View commit details
    Browse the repository at this point in the history

  2. Merge branch 'huggingface:main' into fix-llama-mask-output-attn

    @Aladoro
    Aladoro committed May 6, 2024
    Configuration menu
    Copy the full SHA
    c7bdc95 View commit details
    Browse the repository at this point in the history

  3. Merge branch 'huggingface:main' into fix-llama-mask-output-attn

    @Aladoro
    Aladoro committed May 6, 2024
    Configuration menu
    Copy the full SHA
    8d793a3 View commit details
    Browse the repository at this point in the history

Commits on May 12, 2024

  1. Merge branch 'huggingface:main' into fix-llama-mask-output-attn

    @Aladoro
    Aladoro committed May 12, 2024
    Configuration menu
    Copy the full SHA
    fc143ac View commit details
    Browse the repository at this point in the history

  2. add tests checking sdpa and eager outputs match when output_attention… 

    …s=True
    @Aladoro
    Aladoro committed May 12, 2024
    Configuration menu
    Copy the full SHA
    3e0fada View commit details
    Browse the repository at this point in the history

Commits on May 15, 2024

  1. Split if statements in two lines 

    Co-authored-by: Arthur <[email protected]>
    @Aladoro @ArthurZucker
    Aladoro and ArthurZucker committed May 15, 2024
    Configuration menu
    Copy the full SHA
    9acc119 View commit details
    Browse the repository at this point in the history

  2. Merge branch 'huggingface:main' into fix-llama-mask-output-attn

    @Aladoro
    Aladoro committed May 15, 2024
    Configuration menu
    Copy the full SHA
    08dbd4b View commit details
    Browse the repository at this point in the history

  3. Fix formatting

    @Aladoro
    Aladoro committed May 15, 2024
    Configuration menu
    Copy the full SHA
    9b79aee View commit details
    Browse the repository at this point in the history

  4. Add fix to new jetmoe model

    @Aladoro
    Aladoro committed May 15, 2024
    Configuration menu
    Copy the full SHA
    dd69923 View commit details
    Browse the repository at this point in the history

  5. Add missing output_attentions argument to jetmoe mask creation

    @Aladoro
    Aladoro committed May 15, 2024
    Configuration menu
    Copy the full SHA
    ad4aded View commit details
    Browse the repository at this point in the history