update gpt_model

[Jintao-Huang]

No description provided.

[gemini-code-assist]

Code Review

This pull request refactors the forward methods in GPTModel and MMGPTModel by removing generic **kwargs in favor of more explicit argument handling. A high-severity issue was identified in MMGPTModel where passing the unfiltered kwargs dictionary as extra_block_kwargs could cause a TypeError due to duplicate arguments like input_ids and packed_seq_params. A code suggestion was provided to filter these keys to ensure compatibility.

[gemini-code-assist]

Passing kwargs directly as extra_block_kwargs is problematic because kwargs is mutated at line 85 to include input_ids and packed_seq_params. If self.visual is None, these keys are not removed, leading to a TypeError (multiple values for argument) when calling the language model, as it already receives these arguments explicitly. Additionally, the current logic at line 51 clears kwargs if a visual encoder is present, which would cause any other legitimate extra arguments to be lost. Filtering the dictionary is a safer approach to avoid conflicts with explicit arguments.

Suggested change

	extra_block_kwargs=kwargs,
	extra_block_kwargs={k: v for k, v in kwargs.items() if k not in ['input_ids', 'packed_seq_params']},