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

src/train_model_github_jwjohns.py

@@ -75,7 +75,7 @@ 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)
     # Save original datasets to JSONL
-    with open(f'gsm8k_original_{split}.jsonl', 'w') as f:
+    with open(f'../data/temp/gsm8k_original_{split}.jsonl', 'w') as f:
         for item in data:
             f.write(json.dumps(item) + '\n')
     
@@ -88,7 +88,7 @@ def get_gsm8k_questions(split="train") -> Dataset:
     })
     
     # Save formatted datasets to JSONL
-    with open(f'gsm8k_formatted_{split}.jsonl', 'w') as f:
+    with open(f'../data/temp/gsm8k_formatted_{split}.jsonl', 'w') as f:
         for item in data:
             f.write(json.dumps(item) + '\n')
     return data