[Gemma2] Use nn.SDPA via MultiHeadAttention

[Jack-Khuu]

Gemma2 currently uses a custom attention (Gemma2Attention) instead of the MultiHeadAttention that most models in torchtune use. This is the first of a few PR’s towards making this transition.

What is the purpose of this PR? Is it to

• add a new feature
• fix a bug
• update tests and/or documentation
• other (please add here)

Please link to any issues this PR addresses.

Changelog

What are the changes made in this PR?

• Utilize nn.sdpa via MultiHeadAttention instead of the custom Gemma2Attention.
  • The mask utilized by Gemma2Attention is replaced by get_sliding_attention_mask as a mask_mod to TransformerSelfAttentionLayer.
  • ⚠️ Note: nn.sdpa doesn’t support softcapping (which is a feature of Gemma2). It is reintroduced with FlexAttention support
• Adds scale as an optional arg to the Callable generated from _sdpa_or_flex_attention. This arg is directly exposing the same arg from underlying attention implementation (flex, nn.sdpa)

🚧 This PR does not enable utilizing FlashAttention.
Note: The existing Gemma2Attention implementation will not be deleted until FlexAttention is supported (different PR) and can trivially be swapped back when desired.

Test plan

Please make sure to do each of the following if applicable to your PR. If you're unsure about any one of these just ask and we will happily help. We also have a contributing page for some guidance on contributing.

• run pre-commit hooks and linters (make sure you've first installed via pre-commit install)
• add unit tests for any new functionality
• update docstrings for any new or updated methods or classes
• run unit tests via pytest tests
• run recipe tests via pytest tests -m integration_test
• manually run any new or modified recipes with sufficient proof of correctness
• include relevant commands and any other artifacts in this summary (pastes of loss curves, eval results, etc.)

UX

If your function changed a public API, please add a dummy example of what the user experience will look like when calling it.
Here is a docstring example
and a tutorial example

• I did not change any public API
• I have added an example to docs or docstrings

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchtune/2844

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures, 2 Cancelled Jobs

As of commit 98e2424 with merge base c00aa57 ():

NEW FAILURES - The following jobs have failed:

• GPU tests / gpu_test (3.10, stable) (gh)
  tests/recipes/test_knowledge_distillation_distributed.py::TestKDDistributedRecipe::test_training_state_on_resume_with_async_checkpointing
• Lint / lint (3.10) (gh)

CANCELLED JOBS - The following jobs were cancelled. Please retry:

• GPU tests / gpu_test (3.11, stable) (gh)
  tests/recipes/test_knowledge_distillation_distributed.py::TestKDDistributedRecipe::test_training_state_on_resume_with_async_checkpointing
• GPU tests / gpu_test (3.9, stable) (gh)
  ##[error]The operation was canceled.

This comment was automatically generated by Dr. CI and updates every 15 minutes.