why tokenizer_config.json changed for AWQ model.

#7
by rockcat-miao - opened

compared with deepseek full version, more special token added, chat_template also be changed.
and new chat_template in tokenizer_config.json cannot be Interpreted by vllm tool use parser.

Cognitive Computations org
edited 1 day ago

Because DeepSeek decided that they should update the chat template. This repo uses the old version of said template, I didn't bother to update it. You can just use their newer version though.
Also, my version included a prefill ability, that if a messages list ends with assistant, the model will complete the assistant instead of opening a new response, this is known as prefill.

appreciate your information

rockcat-miao changed discussion status to closed
Your need to confirm your account before you can post a new comment.

Sign up or log in to comment