[Bug for InternVL 4B] ValueError: If `eos_token_id` is defined, make sure that `pad_token_id` is defined. (InternVL-4B)

[yinglang]

model = AutoModel.from_pretrained(
   "OpenGVLab/InternVL2_5-4B",
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    trust_remote_code=True,
    ).eval().to(device)

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)

model.chat(
            tokenizer, pixel_values, question, gen_config,
            num_patches_list=num_patches_list, history=None, return_history=False, verbose=self.verbose)

got Error ValueError: If eos_token_id is defined, make sure that pad_token_id is defined. (InternVL-4B)

[yinglang]

This code solves the problem

llm_generation_config = model.language_model.generation_config
if llm_generation_config.eos_token_id is not None and llm_generation_config.pad_token_id is None: # fixed for InternVL-4B
    llm_generation_config.pad_token_id = llm_generation_config.eos_token_id

model.chat(
            tokenizer, pixel_values, question, gen_config,
            num_patches_list=num_patches_list, history=None, return_history=False, verbose=self.verbose)

I think maybe the problem is due to incrrorect setting in OpenGVLab/InternVL2_5-4B/config.json

 "llm_config": {
    "_name_or_path": "Qwen/Qwen2.5-3B-Instruct",
    "add_cross_attention": false,
    "architectures": [
      "Qwen2ForCausalLM"
    ],
    "eos_token_id": 151645,
    ......
    "pad_token_id": null,

eos_token_id is set but pad_token_id not.

While OpenGVLab/InternVL2_5-8B/config.json is

 "llm_config": {
  "_name_or_path": "internlm/internlm2_5-7b-chat",
  "add_cross_attention": false,
  "architectures": [
    "InternLM2ForCausalLM"
   ],
    "eos_token_id": 2,
    ......
    "pad_token_id": 2,