add md5sum print to model runner

fastdeploy/envs.py

@@ -200,6 +200,9 @@
     "FD_WORKER_ALIVE_TIMEOUT": lambda: int(os.getenv("FD_WORKER_ALIVE_TIMEOUT", "30")),
     # File path for file storage backend
     "FILE_BACKEND_STORAGE_DIR": lambda: str(os.getenv("FILE_BACKEND_STORAGE_DIR", "/tmp/fastdeploy")),
+
+    # Whether to print model's md5 value
+    "MD5SUM_PRINT": lambda: bool(int(os.getenv("MD5SUM_PRINT", "0"))),
 }
 
 

fastdeploy/worker/gpu_model_runner.py

@@ -1399,6 +1399,27 @@
                 )
             else:
                 self.dynamic_weight_manager = DynamicWeightManager(self.fd_config, self.model, self.local_rank)
+
+        # save model md5sum
+        if envs.MD5SUM_PRINT:
+            logger.info("MD5SUM_PRINT is set to True, saving model md5sum...")
+            self._save_model_md5sum()
+
+    def _save_model_md5sum(self) -> None:
+        md5sum_dict = {}
+        for key, param in self.model.state_dict().items():
+            md5sum_dict[key] = get_tensor_md5(param)
+            logger.info(f">>>>>>>>>> {key} : {md5sum_dict[key]}")
+        # 使用 'w' 模式写入
+        file_name = f"md5sum_rank_{paddle.distributed.get_rank()}.json"
+        output_dir = envs.FD_LOG_DIR
+        if not os.path.exists(output_dir):
+            os.makedirs(output_dir, exist_ok=True)
+        with open(os.path.join(output_dir, file_name), "w", encoding="utf-8") as f:
+            # indent=4 让生成的 JSON 有层级缩进，方便肉眼查看
+            # ensure_ascii=False 保证路径中的中文或特殊字符不被转义
+            json.dump(md5sum_dict, f, indent=4, ensure_ascii=False)
+
 
     def get_model(self) -> nn.Layer:
         """Get current model"""