Update app.py

app.py

@@ -4,11 +4,11 @@ import tempfile
 import torch
 import spaces
 import gradio as gr
-from fastapi import FastAPI, UploadFile, File, Form, HTTPException
+from fastapi import UploadFile, File, Form, HTTPException
 from fastapi.responses import JSONResponse, PlainTextResponse
 from transformers import pipeline
 
-# 1. 模型加载
+# 1. 初始化模型
 MODEL_NAME = "openai/whisper-small"
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
@@ -19,9 +19,9 @@ pipe = pipeline(
     device=device
 )
 
-# 2. 核心 GPU 推理函数（ZeroGPU 严格要求必须在文件顶层声明）
+# 2. 【核心】顶层单独定义的 @spaces.GPU 函数（绝对不能被任何类或内部函数包裹）
 @spaces.GPU
-def run_whisper_inference(audio_path: str, target_language: str = None, is_translate: bool = False):
+def run_whisper(audio_path: str, target_language: str = None, is_translate: bool = False):
     generate_kwargs = {}
     if target_language:
         generate_kwargs["language"] = target_language
@@ -33,75 +33,55 @@ def run_whisper_inference(audio_path: str, target_language: str = None, is_trans
     result = pipe(audio_path, generate_kwargs=generate_kwargs)
     return result["text"]
 
-# 3. Gradio UI 封装
+
+# 3. 构建 Gradio 界面
 def gradio_predict(audio_path):
     if audio_path is None:
-        return "请先上传音频或录音！"
-    try:
-        return run_whisper_inference(audio_path)
-    except Exception as e:
-        return f"错误: {str(e)}"
+        return "请上传音频文件！"
+    return run_whisper(audio_path)
 
 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】"
+    outputs=gr.Textbox(label="识别结果"),
+    title="Whisper API Node"
 )
 
-# 4. FastAPI 应用声明
-app = FastAPI()
+# 4. 直接使用 Gradio 自带的 demo.app（避免零 GPU 扫描机制找不到路由）
+app = demo.app
 
-async def process_openai_audio_request(file: UploadFile, response_format: str, language: str, is_translate: bool):
+async def process_audio(file: UploadFile, response_format: str, language: str, is_translate: bool):
     suffix = os.path.splitext(file.filename)[1] or ".wav"
     with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as temp_file:
         shutil.copyfileobj(file.file, temp_file)
         temp_path = temp_file.name
 
     try:
-        text = run_whisper_inference(temp_path, target_language=language, is_translate=is_translate)
+        text = run_whisper(temp_path, target_language=language, is_translate=is_translate)
     except Exception as e:
-        raise HTTPException(status_code=500, detail=f"Inference failed: {str(e)}")
+        raise HTTPException(status_code=500, detail=str(e))
     finally:
         if os.path.exists(temp_path):
             os.remove(temp_path)
 
     if response_format in ["text", "vtt", "srt"]:
         return PlainTextResponse(text)
-
     return JSONResponse(content={"text": text})
 
+# 5. 添加 OpenAI 兼容接口
 @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)
+    response_format: str = Form("json")
 ):
-    return await process_openai_audio_request(
-        file=file,
-        response_format=response_format,
-        language=language,
-        is_translate=False
-    )
+    return await process_audio(file, response_format, language, is_translate=False)
 
 @app.post("/v1/audio/translations")
 async def translate_api(
     file: UploadFile = File(...),
     model: str = Form("whisper-1"),
-    prompt: str = Form(None),
-    response_format: str = Form("json"),
-    temperature: float = Form(0.0)
+    response_format: str = Form("json")
 ):
-    return await process_openai_audio_request(
-        file=file,
-        response_format=response_format,
-        language="english",
-        is_translate=True
-    )
-
-# 5. 挂载 Gradio 到 FastAPI
-app = gr.mount_gradio_app(app, demo, path="/")
+    return await process_audio(file, response_format, language="english", is_translate=True)