fix data form support

handler.py

@@ -34,6 +34,9 @@ class EndpointHandler:
             "prompt": "描述图片内容的自然语言指令"
         }
         """
+        # Hugging Face接口会把真正的输入放在inputs字段里
+        data = data.get("inputs", data)
+
         # 判断输入模态
         if "video" in data:
             modal = "video"