fix #2125

requirements.txt

@@ -6,6 +6,7 @@ peft>=0.7.0
 trl>=0.7.6
 gradio>=3.38.0,<4.0.0
 scipy
+einops
 sentencepiece
 protobuf
 tiktoken
@@ -17,4 +18,3 @@ pydantic
 fastapi
 sse-starlette
 matplotlib
-einops

src/llmtuner/webui/runner.py

@@ -118,13 +118,13 @@ class Runner:
             logging_steps=get("train.logging_steps"),
             save_steps=get("train.save_steps"),
             warmup_steps=get("train.warmup_steps"),
-            neftune_noise_alpha=get("train.neftune_alpha"),
+            neftune_noise_alpha=get("train.neftune_alpha") or None,
             train_on_prompt=get("train.train_on_prompt"),
             upcast_layernorm=get("train.upcast_layernorm"),
             lora_rank=get("train.lora_rank"),
             lora_dropout=get("train.lora_dropout"),
             lora_target=get("train.lora_target") or get_module(get("top.model_name")),
-            additional_target=get("train.additional_target") if get("train.additional_target") else None,
+            additional_target=get("train.additional_target") or None,
             create_new_adapter=get("train.create_new_adapter"),
             output_dir=get_save_dir(get("top.model_name"), get("top.finetuning_type"), get("train.output_dir"))
         )
@@ -164,7 +164,6 @@ class Runner:
 
         args = dict(
             stage="sft",
-            do_eval=True,
             model_name_or_path=get("top.model_path"),
             adapter_name_or_path=adapter_name_or_path,
             cache_dir=user_config.get("cache_dir", None),
@@ -187,8 +186,9 @@ class Runner:
         )
 
         if get("eval.predict"):
-            args.pop("do_eval", None)
             args["do_predict"] = True
+        else:
+            args["do_eval"] = True
 
         return args