tie_word_embeddings为true的模型，LoRA 微调，additional_target只学习 embed_tokens 时，推理时能够输出新添加的特殊 Token，但同时学习 embed_tokens 和 lm_head 后则不会 #1899
Triggered via issue
December 25, 2024 09:38
huangxuyao
Status
Success
Total duration
11s
Artifacts
–
Annotations
1 warning
|
label_issue
ubuntu-latest pipelines will use ubuntu-24.04 soon. For more details, see https://github.com/actions/runner-images/issues/10636
|