new

app.py

@@ -30,66 +30,72 @@ except Exception as e:
     print(f"Error initializing OpenAI client: {str(e)}")
     raise
 
+# 设置环境变量以禁用 CUDA 警告
+os.environ['CUDA_VISIBLE_DEVICES'] = ''
+
 # 设置模型下载目录
 MODEL_CACHE_DIR = os.path.join(os.path.dirname(__file__), 'models')
 os.makedirs(MODEL_CACHE_DIR, exist_ok=True)
 
-# 检查 GPU 环境
+# 简化 GPU 检查
 def check_gpu():
     try:
         if torch.cuda.is_available():
-            # 获取 GPU 信息
-            gpu_name = torch.cuda.get_device_name(0)
-            gpu_memory = torch.cuda.get_device_properties(0).total_memory / 1024**3  # 转换为 GB
-            print(f"Found GPU: {gpu_name} with {gpu_memory:.2f}GB memory")
             return 'cuda'
-        else:
-            print("No CUDA GPU available")
-            return 'cpu'
-    except Exception as e:
-        print(f"Error checking GPU: {str(e)}")
-        return 'cpu'
+    except:
+        pass
+    return 'cpu'
 
 # 初始化设备
 device = check_gpu()
 print(f"Running on device: {device}")
 
-# 初始化 EasyOCR（针对 T4 GPU 优化）
-def initialize_easyocr():
+# 预下载模型
+def download_models():
     try:
-        print("Initializing EasyOCR and loading models...")
-        if device == 'cuda':
-            # 为 T4 GPU 设置较小的批处理大小和内存限制
-            torch.cuda.empty_cache()  # 清理 GPU 内存
-            reader = easyocr.Reader(
-                ['ch_sim', 'en'],
-                gpu=True,
-                download_enabled=True,
-                verbose=True,
-                model_storage_directory=MODEL_CACHE_DIR,
-                recog_batch_size=8,  # 减小批处理大小
-                detector_batch_size=2
-            )
-        else:
-            reader = easyocr.Reader(
+        print("Checking for pre-downloaded models...")
+        model_files = [
+            os.path.join(MODEL_CACHE_DIR, 'craft_mlt_25k.pth'),
+            os.path.join(MODEL_CACHE_DIR, 'chinese_sim.pth'),
+            os.path.join(MODEL_CACHE_DIR, 'english_g2.pth')
+        ]
+        
+        all_models_exist = all(os.path.exists(f) for f in model_files)
+        if not all_models_exist:
+            print("Some models need to be downloaded...")
+            # 强制在 CPU 模式下下载模型
+            temp_reader = easyocr.Reader(
                 ['ch_sim', 'en'],
                 gpu=False,
+                model_storage_directory=MODEL_CACHE_DIR,
                 download_enabled=True,
-                verbose=True,
-                model_storage_directory=MODEL_CACHE_DIR
+                verbose=True
             )
+            print("Model download completed")
+        else:
+            print("All models already downloaded")
+    except Exception as e:
+        print(f"Error during model download: {str(e)}")
+
+# 下载模型
+download_models()
+
+# 初始化 EasyOCR
+def initialize_easyocr():
+    try:
+        print("Initializing EasyOCR...")
+        reader = easyocr.Reader(
+            ['ch_sim', 'en'],
+            gpu=False,  # 强制使用 CPU 模式
+            model_storage_directory=MODEL_CACHE_DIR,
+            download_enabled=False,  # 禁用自动下载
+            verbose=True
+        )
         print("EasyOCR initialization completed!")
         return reader
     except Exception as e:
         print(f"Error initializing EasyOCR: {str(e)}")
-        print("Falling back to CPU mode...")
-        return easyocr.Reader(
-            ['ch_sim', 'en'],
-            gpu=False,
-            download_enabled=True,
-            verbose=True,
-            model_storage_directory=MODEL_CACHE_DIR
-        )
+        raise
 
 # 初始化 reader
 reader = initialize_easyocr()
@@ -124,15 +130,10 @@ def extract_text_from_image(image):
             image.save(image_path)
         
         print("开始识别文字...")
-        if device == 'cuda':
-            torch.cuda.empty_cache()  # 清理 GPU 内存
-        
-        # 使用 EasyOCR 识别文字
         result = reader.readtext(
             image_path,
             detail=1,
-            paragraph=True,
-            batch_size=8  # 控制批处理大小
+            paragraph=True
         )
         print("文字识别完成")
         
@@ -252,8 +253,7 @@ with gr.Blocks(title="课程幻灯片理解助手") as demo:
     if api_key_error:
         gr.Markdown(api_key_error)
     else:
-        gpu_info = f"GPU (T4)" if device == 'cuda' else "CPU"
-        gr.Markdown(f"# 📚 课程幻灯片理解助手 ({gpu_info} 模式)")
+        gr.Markdown("# 📚 课程幻灯片理解助手")
         gr.Markdown("上传幻灯片图片，AI 将自动识别内容并提供详细讲解")
         
         # 存储当前识别的文字，用于对话上下文
@@ -333,5 +333,8 @@ with gr.Blocks(title="课程幻灯片理解助手") as demo:
 
 # 启动应用
 if __name__ == "__main__":
-    # 设置较小的并行处理数
-    demo.launch(share=True, max_threads=4)
+    demo.launch(
+        share=True,
+        max_threads=4,
+        show_error=True
+    )

requirements.txt

@@ -5,5 +5,4 @@ python-dotenv>=1.0.0
 openai>=1.0.0
 Pillow>=10.0.0
 numpy>=1.24.0
-torch>=2.0.0
-torchvision>=0.15.0
+torch>=2.0.0