瀏覽代碼

修复ppo训练报错

lxylxy123321 21 小時之前
父節點
當前提交
5980af33a7
共有 1 個文件被更改,包括 1 次插入1 次删除
  1. 1 1
      backend/app/engines/text_engine.py

+ 1 - 1
backend/app/engines/text_engine.py

@@ -333,7 +333,7 @@ class TextEngine(BaseEngine):
                 gradient_accumulation_steps=gradient_accumulation,
                 ppo_epochs=ppo_epochs,
                 vf_coef=vf_coef,
-                kl_ctl=kl_coef,
+                init_kl_coef=kl_coef,
                 response_length=response_length,
                 output_dir=output_dir,
                 logging_steps=10,