Qwen2.5 icon indicating copy to clipboard operation
Qwen2.5 copied to clipboard

Qwen2-7B 输出结果一直重复,只到max_new_tokens长度

Open shilei4260 opened this issue 7 months ago • 4 comments

我用的是Qwen2-7B 不是Qwen2-7B-Instruct 因为Qwen2-7B 是128k ,Qwen2-7B-Instruct 32k 用vllm推理Qwen2-7B默认的参数会一直重复直到max_new_tokens的长度,我修改tokenizer_config.json中eos_token为<|im_end|>后有时候还是会出现一直重复直到max_new_tokens的长度。 {'messages': [{'role': 'user', 'content': '1+1'}], 'model': 'Qwen2-7B', "stream":true,} Uploading 捕获2.JPG…

shilei4260 avatar Jul 19 '24 04:07 shilei4260