ssbuild / chatglm2_finetuning

chatglm2 6b finetuning and alpaca finetuning
Apache License 2.0
144 stars 17 forks source link

lora 训练完成后推断回复为空 #31

Closed Derican closed 1 year ago

Derican commented 1 year ago

lora训练完成后用infer_lora_finetuning.py推断,显示

============================================================ lora info
trainable params: 0 || all params: 6,245,533,696 || trainable%: 0.0

并且response为空,请问是哪里的config没有设置还是训练没有配置好

ssbuild commented 1 year ago

需要调参,例如增大lora rank , 增大训练epoch 等等

Derican commented 1 year ago

明白了,换了个优化器解决了。