换用github jwjohns/unsloth-GRPO-qwen2.5 验证GRPO训练模型

src/train_model_github_jwjohns.py

@@ -74,6 +74,7 @@ from modelscope.msdatasets import MsDataset
 def get_gsm8k_questions(split="train") -> Dataset:
     # data = load_dataset('openai/gsm8k', 'main')[split]
     data =  MsDataset.load('openai-mirror/gsm8k', subset_name='main', split=split)
+    os.makedirs(f'../data/temp/',exist_ok=False)
     # Save original datasets to JSONL
     with open(f'../data/temp/gsm8k_original_{split}.jsonl', 'w') as f:
         for item in data: