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

src/train_model_github_jwjohns.py

@@ -227,6 +227,8 @@ trainer.train()
 
 # Save the trained model
 print("Saving LoRA weights to grpo_saved_lora...")
-model.save_lora(f"../models/trained/grpoModel")
+# model.save_lora(f"../models/trained/grpoModel")
+model.save_pretrained(f"../models/trained/grpoModel")
+tokenizer.save_pretrained(f"../models/trained/grpoModel")
 
 print("Training complete!")