[qwen3.5] compat transformers 5.4.0

[Jintao-Huang]

No description provided.

[gemini-code-assist]

Code Review

This pull request introduces an is_mtp flag to several methods in gpt_bridge.py to correctly handle Multi-Token Prediction (MTP) layers during model conversion, particularly for MoE architectures. It also adds a safety check for labels in the _postprocess method of gpt_model.py. The review feedback identifies an improvement in _get_hf_experts_attr to better support multimodal models using Qwen 3.5 MoE and to ensure MTP layers correctly fall through the logic by removing redundant entries.