xiaobo-Chen
Results
1
issues of
xiaobo-Chen
### Reminder - [X] I have read the README and searched the existing issues. ### Reproduction 对qwen1.5-7b-chat模型进行DPO训练,由于输入长度较长,指定cutoff_len=8192。根据https://github.com/hiyouga/LLaMA-Factory/blob/85c376fc1e0bcc854ed6e70e6455a0b00b341655/src/llmtuner/data/preprocess.py#L27来看应该是把input 进行tokenizer之后再对id进行cutoff。使用chat对应的Template中存在 ,是否也会进行截断,进而影响训练效果?是否能作为一个优化方向? ### Expected behavior _No response_ ### System Info _No response_ ###...
duplicate