Improve _update_causal_mask performance

[alessandropalla]

What does this PR do?

Fixes # (issue)
#29206

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@ArthurZucker

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[ArthurZucker]

Thanks! Is there other changes you need to put in?

[alessandropalla]

@ArthurZucker I think it can be merged, if I found additional optimizations I'll add them into another PR

[ArthurZucker]

LGTM, can you run the slow tests for both models? (otherwise I'll spin an instance and do it)

[alessandropalla]

So, just ran locally tests/models/[gemma,llama]/test_modeling_[gemma,llama].py and got a bunch of errors that I'm not sure are related to the PR but most probably from my environment

FAILED tests/models/gemma/test_modeling_gemma.py::GemmaModelTest::test_cpu_offload - AssertionError: Items in the second set but not the first:
FAILED tests/models/gemma/test_modeling_gemma.py::GemmaModelTest::test_disk_offload_bin - ValueError: You are trying to offload the whole model to the disk. Please use the `disk_offload` function instead.
FAILED tests/models/gemma/test_modeling_gemma.py::GemmaModelTest::test_disk_offload_safetensors - ValueError: You are trying to offload the whole model to the disk. Please use the `disk_offload` function instead.
FAILED tests/models/gemma/test_modeling_gemma.py::GemmaModelTest::test_model_parallelism - AssertionError: Items in the second set but not the first:

FAILED tests/models/llama/test_modeling_llama.py::LlamaModelTest::test_cpu_offload - AssertionError: Items in the second set but not the first:
FAILED tests/models/llama/test_modeling_llama.py::LlamaModelTest::test_disk_offload_bin - ValueError: You are trying to offload the whole model to the disk. Please use the `disk_offload` function instead.
FAILED tests/models/llama/test_modeling_llama.py::LlamaModelTest::test_disk_offload_safetensors - ValueError: You are trying to offload the whole model to the disk. Please use the `disk_offload` function instead.
FAILED tests/models/llama/test_modeling_llama.py::LlamaModelTest::test_model_parallelism - AssertionError: Items in the second set but not the first:

In my env, set(new_model.hf_device_map.values()) returns only {cpu}, also the other error seems to be device_map related

[ArthurZucker]

Yes those are unrelated!

[ArthurZucker]

Thanks for the PR 🤗