Update app.py

app.py

@@ -1,6 +1,5 @@
 import torch
 from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
-from transformers import pipeline
 from datasets import load_dataset
 import spacy
 import gradio as gr
@@ -11,19 +10,29 @@ torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
 
 # Whisper 模型初始化（語音轉文字）
 whisper_model_id = "openai/whisper-large-v3"
-whisper_model = AutoModelForSpeechSeq2Seq.from_pretrained(
-    whisper_model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
-)
-whisper_model.to(device)
-whisper_processor = AutoProcessor.from_pretrained(whisper_model_id)
 
-whisper_pipe = pipeline(
-    "automatic-speech-recognition",
-    model=whisper_model,
-    tokenizer=whisper_processor.tokenizer,
-    feature_extractor=whisper_processor.feature_extractor,
-    device=device,
-)
+try:
+    whisper_model = AutoModelForSpeechSeq2Seq.from_pretrained(
+        whisper_model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True,
+        revision="main",  # 嘗試指定修訂版以解決兼容性問題
+    )
+    whisper_model.to(device)
+    
+    # 加載處理器時也指定修訂版（如果需要）
+    whisper_processor = AutoProcessor.from_pretrained(whisper_model_id)
+except Exception as e:
+    print(f"加載Whisper模型或處理器時出現錯誤：{e}")
+else:
+    # 成功加載後建立pipeline進行語音轉文字工作
+    
+    whisper_pipe = pipeline(
+        "automatic-speech-recognition",
+        model=whisper_model,
+        tokenizer=whisper_processor.tokenizer,
+        feature_extractor=whisper_processor.feature_extractor,
+        device=device,
+        
+    )
 
 # DeepSeek-V3 模型初始化（文本生成）
 deepseek_pipe = None  # 預設值，以防模型加載失敗
@@ -31,56 +40,64 @@ deepseek_pipe = None  # 預設值，以防模型加載失敗
 try:
     deepseek_pipe = pipeline("text-generation", model="deepseek-ai/DeepSeek-V3", trust_remote_code=True)
 except Exception as e:
-    print(f"加載模型時出現錯誤：{e}")
+    print(f"加載DeepSeek-V3模型時出現錯誤：{e}")
 
 # spaCy 初始化（文本分類與標籤）
 nlp = spacy.load("en_core_web_sm")
 
 def process_audio(audio_file):
-    # 語音轉文字
-    result = whisper_pipe(audio_file)["text"]
     
-    # 使用 DeepSeek 生成回應（如果成功加載模型）
-    if deepseek_pipe is not None:
-        messages = [{"role": "user", "content": result}]
-        deepseek_response = deepseek_pipe(messages)[0]["generated_text"]
+    try:
         
-        # 使用 spaCy 分析文本
-        doc = nlp(deepseek_response)
-        entities = [(ent.text, ent.label_) for ent in doc.ents]
+        result = whisper_pipe(audio_file)["text"]
         
-        return {
-            "Transcription (Whisper)": result,
-            "AI Response (DeepSeek)": deepseek_response,
-            "Extracted Entities (spaCy)": entities,
             
-            }
-    
         
             
-    
+            # 使用 DeepSeek 生成回應（如果成功加載模型）
+            if deepseek_pipe is not None:
+                messages=[{"role": "user", "content": result}]
+                deepseek_response=deepseek_pipe(messages)[0]["generated_text"]
+                
+                doc=nlp(deepseek_response)
+                entities=[(ent.text, ent.label_) for ent in doc.ents]
+                
+                return {
+                    "Transcription (Whisper)": result,
+                    "AI Response (DeepSeek)": deepseek_response,
+                    "Extracted Entities (spaCy)": entities
+                    
+                    
+                    
+                    }
+                
+            
                 
             
         
-    
+        
     
     
 
 with gr.Blocks() as app:
+
+   
+   
    
    with gr.Row():
        audio_input=gr.Audio(source="microphone", type="filepath", label="上傳語音")
        output_text=gr.JSON(label="結果")
        
-   submit_button=gr.Button("提交")  # 修正了這一行
-   
-   
-   
-   
-   
+       
+       
+       
+       
+       
+       
   
   
 
+submit_button=gr.Button("提交") 
 submit_button.click(fn=lambda x: process_audio(x), inputs=[audio_input], outputs=[output_text])
 
 if __name__ == "__main__":