diff --git a/src/evaluation.py b/src/evaluation.py
index d28ea68..be5b01e 100644
--- a/src/evaluation.py
+++ b/src/evaluation.py
@@ -9,11 +9,11 @@ from trl import (
     get_quantization_config,
 )
 
-from utils.args import ContinualScriptArguments
+from utils.args import ContinualScriptArguments, ContinualModelConfig
 
 
 if __name__ == "__main__":
-    parser = TrlParser((ContinualScriptArguments, TrainingArguments, ModelConfig))
+    parser = TrlParser((ContinualScriptArguments, TrainingArguments, ContinualModelConfig))
     script_args, training_args, model_args = parser.parse_args_and_config()
     # for type hint
     if 0 == 1:
diff --git a/src/evaluation.sh b/src/evaluation.sh
index e50bf72..a440e82 100755
--- a/src/evaluation.sh
+++ b/src/evaluation.sh
@@ -1,8 +1,9 @@
 #!/bin/bash
 
-accelerate launch --config_file accelerate_configs/deepspeed_zero2.yaml evaluation.py \
+accelerate launch --config_file configs/accelerate_configs/deepspeed_zero2.yaml evaluation.py \
     --dataset_name OCR_VQA_200K \
     --use_peft \
+    --peft_type MMOELORA \
     --model_name_or_path Qwen/Qwen2-VL-7B-Instruct \
     --lora_target_modules q_proj v_proj \
     --per_device_train_batch_size 1 \
diff --git a/src/todo.md b/src/todo.md
index 4538288..8e7bf4a 100644
--- a/src/todo.md
+++ b/src/todo.md
@@ -11,5 +11,6 @@
 
 [2025.01.03]
 
-- [ ] 处理量化逻辑（懒得调了）
+- [ ] 处理量化逻辑
 - [ ] 严查moelora的原始代码，太粗糙了😡
+- [ ] 未知原因trainer后处理时间长
diff --git a/src/train.py b/src/train.py
index d9a276c..f2d3c2f 100644
--- a/src/train.py
+++ b/src/train.py
@@ -114,7 +114,13 @@ if __name__ == "__main__":
         )
         trainer.train()
 
+        if accelerator.is_local_main_process:
+            print("Saving model")
         trainer.save_model(training_args.output_dir)
+        if accelerator.is_local_main_process:
+            print("Model saved")
+        # 同步 accelerator
+        accelerator.wait_for_everyone()
 
         model.eval()
         accelerator = trainer.accelerator