fix models for transformers>=5

[grimoire]

Fix failed model for transformers>=5

Gemma3 force linear rope.

[Copilot]

Pull request overview

This PR updates several VLM/model integration points to restore compatibility with newer Hugging Face Transformers (>=5), including config nesting changes (e.g., text_config) and updated processor/model configuration fields.

Changes:

• Adjust Qwen2-VL/Qwen2.5-VL PyTorch model construction to use text_config for LLM components (model + LM head).
• Update Qwen2-VL image preprocessing call signature to align with newer image_processor APIs.
• Add a dedicated Qwen2VLModelConfigBuilder and harden a few config reads (bos_token_id, Gemma hidden activation / RoPE handling).

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
lmdeploy/vl/model/qwen2.py	Removes videos=None arg from image_processor call for newer processor signatures.
lmdeploy/vl/model/cogvlm.py	Pins Vicuna eoa to </s> when building an internal chat template.
lmdeploy/pytorch/models/qwen2_vl.py	Uses config.text_config (when present) for building the text model + lm_head.
lmdeploy/pytorch/models/qwen2_5_vl.py	Same text_config handling for Qwen2.5-VL.
lmdeploy/pytorch/models/llama4.py	Removes unused pad_token_id attribute assignment.
lmdeploy/pytorch/models/gemma.py	Makes Gemma activation/rope construction more tolerant of config differences, adds Gemma3-specific RoPE building logic.
lmdeploy/pytorch/configurations/qwen2_vl.py	New config builder to route Qwen2-VL configs through text_config and propagate quantization/dtype.
lmdeploy/pytorch/configurations/deepseek_v2.py	Avoids attribute errors by reading bos_token_id via getattr.
lmdeploy/pytorch/configurations/chatglm.py	Same bos_token_id hardening via getattr.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[BBC-Esq]

You'll also need to ensure that llm-compressor updates it's library as well since it's still requiring lower versions of transformers and other libraries...

[lvhan028]

Hi @BBC-Esq, thanks for pointing this out. We'll address the lmdeploy-lite-package-compatible-transformers-5 issue in a follow-up PR