Update pretrain_gpt.py 添加全量编译

pretrain_gpt.py

@@ -128,7 +128,7 @@ def model_provider(pre_process=True, post_process=True) -> Union[GPTModel, megat
                 rotary_base=args.rotary_base,
                 rope_scaling=args.use_rope_scaling
             )
-    #model = torch.compile(model,mode='max-autotune-no-cudagraphs')
+    model = torch.compile(model,mode='max-autotune-no-cudagraphs')
     print_rank_0(model)
 
     return model