Update app.py

app.py

@@ -10,9 +10,7 @@ from transformers import pipeline
 # 1. 声明加载的模型
 MODEL_NAME = "openai/whisper-small"
 
-# 2. 全局直接在 "cuda" 设备上初始化 Pipeline！
-# 官方的 spaces 拦截层会在容器启动时自动拦截它防止在 CPU 阶段报错；
-# 并在调用带 @spaces.GPU 装饰器的函数时，自动、无损地把它调度到 A100 GPU 显存中运行。
+# 2. 全局初始化 Pipeline
 pipe = pipeline(
     "automatic-speech-recognition",
     model=MODEL_NAME,
@@ -20,20 +18,17 @@ pipe = pipeline(
     device="cuda"
 )
 
-# 3. 核心计算函数（保留 @spaces.GPU 装饰器，确保 Hugging Face 静态扫描通过）
+# 3. 核心计算函数
 @spaces.GPU
 def transcribe_core(audio_path: str, target_language: str = None, is_translate: bool = False):
-    # 【最关键的安全操作】绝不手动编写 pipe.model.to("cuda") 或 torch.autocast
-    # 彻底杜绝由于显卡热插拔带来的 RuntimeError！
     generate_kwargs = {}
     if target_language:
         generate_kwargs["language"] = target_language
-        
+
     if is_translate:
         generate_kwargs["language"] = "english"
         generate_kwargs["task"] = "translate"
-        
-    # 直接运行推理，ZeroGPU 机制会完美托管这一步
+
     result = pipe(audio_path, generate_kwargs=generate_kwargs)
     return result["text"]
 
@@ -44,19 +39,22 @@ def gradio_predict(audio_path):
     try:
         return transcribe_core(audio_path)
     except Exception as e:
-        return f"转录出错: {str(e)}"
+        return f"错误: {str(e)}"
 
 demo = gr.Interface(
     fn=gradio_predict,
     inputs=gr.Audio(sources=["microphone", "upload"], type="filepath", label="输入音频"),
     outputs=gr.Textbox(label="识别出的文本"),
     title="Whisper 语音识别 API 节点",
-    description="【完美兼容 OpenAI 规范 - A100 GPU 版】支持网页端测试，同时提供 100% 兼容的 /v1/audio/transcriptions & /v1/audio/translations 接口！"
+    description="【完美兼容 OpenAI 规范】"
 )
 
-app = demo.app
+# ===========================================================================
+# 关键修改部分：正确初始化 FastAPI 应用
+# ===========================================================================
+app = FastAPI()
 
-# --- 完美兼容 OpenAI 的处理函数 ---
+# 辅助函数：处理 OpenAI 请求
 async def process_openai_audio_request(file, response_format, language, is_translate):
     suffix = os.path.splitext(file.filename)[1] or ".mp3"
     with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as temp_file:
@@ -73,7 +71,7 @@ async def process_openai_audio_request(file, response_format, language, is_trans
 
     if response_format in ["text", "vtt", "srt"]:
         return PlainTextResponse(text)
-    
+
     return JSONResponse(content={"text": text})
 
 
@@ -81,11 +79,11 @@ async def process_openai_audio_request(file, response_format, language, is_trans
 @app.post("/v1/audio/transcriptions")
 async def transcribe_api(
     file: UploadFile = File(...),
-    model: str = Form("whisper-1"),           
-    language: str = Form(None),               
-    prompt: str = Form(None),                 
-    response_format: str = Form("json"),       
-    temperature: float = Form(0.0)            
+    model: str = Form("whisper-1"),
+    language: str = Form(None),
+    prompt: str = Form(None),
+    response_format: str = Form("json"),
+    temperature: float = Form(0.0)
 ):
     return await process_openai_audio_request(
         file=file,
@@ -111,5 +109,5 @@ async def translate_api(
         is_translate=True
     )
 
-if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", server_port=7860)
+# 6. 将 Gradio 挂载到 FastAPI 应用的根路径（必须放在所有 FastAPI 路由之后！）
+app = gr.mount_gradio_app(app, demo, path="/")