[BUG] <Response from Qwen-VL very short>

[zifuwan]

是否已有关于该错误的issue或讨论？ | Is there an existing issue / discussion for this?

• 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions

该问题是否在FAQ中有解答？ | Is there an existing answer for this in FAQ?

• 我已经搜索过FAQ | I have searched FAQ

当前行为 | Current Behavior

When I generate an answer to question "Please describe the image", the answer is very short (probably just 2 or 3 words).
Here's the code for my generation:

output_ids = model.generate(
                    input_ids=input_ids.input_ids.cuda(),
                    attention_mask=input_ids.attention_mask.cuda(),
                    images=image.unsqueeze(0).half().cuda(),
                    images_pos=(image_pos.unsqueeze(0).half().cuda() if image_pos is not None else None),
                    images_neg=(image_neg.unsqueeze(0).half().cuda() if image_neg is not None else None),
                    do_sample=False,
                    temperature=args.temperature,
                    top_p=args.top_p,
                    top_k=args.top_k,
                    max_new_tokens=args.max_new_tokens,
                    num_beams=1,
                    min_new_tokens=1,
                    length_penalty=1,
                    num_return_sequences=1,
                    output_hidden_states=True,
                    use_cache=True,)
outputs = [
            tokenizer.decode(_[input_ids.input_ids.size(1):].cpu(),
                             skip_special_tokens=True).strip() for _ in output_ids
        ][0]
        outputs = outputs.strip()

期望行为 | Expected Behavior

Hope to get a longer response.

复现方法 | Steps To Reproduce

No response

运行环境 | Environment

- OS:
- Python:
- Transformers:
- PyTorch:
- CUDA (`python -c 'import torch; print(torch.version.cuda)'`):

备注 | Anything else?

No response