xtuner
xtuner copied to clipboard
请问目前支持qwen2吗?
我看文档里只写支持到qwen1.5,但是issue里不少人有用在qwen2上?
我想在qwen2上用序列并行训长文本
Sequence parallel needs transformers <4.43. Same issue in #935
Sequence parallel needs transformers <4.43. Same issue in #935
训了一版,不过loss看着不太正常,性能也没提升