qwen72b-chat扩展到32k需要修改哪些参数呢?

#3
by TianyuLLM - opened

是只需要修改generation_config.json中的max_window_size参数吗?
我只修改了这个参数,在使用长文本>8k时,会报警告:token indices sequence length is longer than the specified maximum sequence length for this model (12805>8192).
需要修改其它参数吗?

token indices sequence length is longer than the specified maximum sequence length for this model (12805>8192)

这个没关系的哈,忽略就好。嫌它烦人可以改下tokenizer_config.json里的max_model_length (因为tokenizer还有pad和truncate的功能需要这个配置项,一般不会用到,但得有…)

其实默认配置应该是就可以8K的,其它参数还有max_new_tokens按需配置就可以。

jklj077 changed discussion status to closed

Sign up or log in to comment