Adapt transformers v4.57.0

[Kaihui-intel]

Type of Change

bug fix

Description

From transformers huggingface/transformers#41103:
completely remove offload_state_dict: it was added several years ago to avoid loading 2x memory on cpu between the model and the state dict -> this is no longer needed from some time as the model is loaded on meta device, then params are loaded one after the other -> removes quite a bit of convoluted logic

Expected Behavior & Potential Risk

the expected behavior that triggered by this PR

How has this PR been tested?

how to reproduce the test (including hardware information)

Dependency Change?

any library dependency introduced or removed

[PRAgent4INC]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Code Order The new condition for adding offload_state_dict is placed after the check for transformers version less than 4.51. This might lead to incorrect behavior if offload_state_dict should be set for versions between 4.51 and 4.57. if parse(transformers.__version__) < parse("4.57"): tmp_kwargs["offload_state_dict"] = offload_state_dict

[PRAgent4INC]

PR Code Suggestions ✨

No code suggestions found for the PR.

[Kaihui-intel]

AutoRound CI will fix in another PR #2291