Closed linhaojia13 closed 1 month ago
The easiest way is to change <|endoftext|>
to <|im_end|>
here. However, conv_bunny
is used for phi-1.5, phi-2, stalelm-2 and qwen1.5, so this change would invalidate the compatibility with other models.
You may define a new conv_qwen_chat
and pay attention to all usages like conv_mode
and version
.
Thank you very much!
Qwen1.5-1.8B/config.json:
Qwen1.5-1.8B-Chat/config.json:
This difference cause the condition
if tokenizer.pad_token_id == tokenizer.eos_token_id:
inpreprocess_bunny
is different, which can be seen as follow:If I want to modify
preprocess_bunny
forqwen-1.5-1.8b-chat
, it seems that there are two ways: 1) deleteround_len += 1
; or 2) deleteif tokenizer.pad_token_id == tokenizer.eos_token_id
and makecur_len -= end_token_cnt
always be done.I'm not sure which way is correct and will not bring potential errors.