423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-10-14 07:42:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

[data] fix reasoning template (#9219)

Browse Source
This commit is contained in:
Yaowei Zheng 2025-09-30 18:11:45 +08:00 committed by GitHub
parent 09dedf144f
commit e2b1594d31
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/llamafactory/data/template.py
View File

@ -416,8 +416,8 @@ class ReasoningTemplate(Template):
prompt_ids, response_ids = super().encode_oneturn(tokenizer, messages, system, tools)
if (
self.thought_words[0] not in messages[-1]["content"]
and self.thought_words[1] not in messages[-1]["content"]
self.thought_words[0].strip() not in messages[-1]["content"]
and self.thought_words[1].strip() not in messages[-1]["content"]
): # add empty cot
if not self.enable_thinking: # do not compute loss
prompt_ids += self.get_thought_word_ids(tokenizer)
@ -442,8 +442,8 @@ class ReasoningTemplate(Template):
encoded_messages = self._encode(tokenizer, messages, system, tools)
for i in range(0, len(messages), 2):
if (
self.thought_words[0] not in messages[i + 1]["content"]
and self.thought_words[1] not in messages[i + 1]["content"]
self.thought_words[0].strip() not in messages[i + 1]["content"]
and self.thought_words[1].strip() not in messages[i + 1]["content"]
): # add empty cot
if not self.enable_thinking: # do not compute loss
encoded_messages[i] += self.get_thought_word_ids(tokenizer)