qwen1.5-7b-chat是不是推理起来比qwen1.5-7b快很多
#9
by
endNone
- opened
跑fastchat/mtbench的时候,我发现qwen1.5-7b-chat花费的时间是qwen1.5-7b不到一半,是官方对qwen1.5-7b-chat做了什么加速处理吗
There are no difference in code. But since MTBench is for evaluating chat models, my guess it that the base model qwen1.5-7b generated overly long responses and thus cost more time.
I have compared the length of the generated dialogues, and the chat model produces longer ones.
This comment has been hidden
JustinLin610
changed discussion status to
closed