Update app.py

app.py

@@ -68,7 +68,7 @@ class TextToSQLSystem:
         self._log("初始化系統...")
         self.query_cache = {}
 
-        # 1. 載入嵌入模型（使用 transformers）
+        # 1. 載入嵌入模型
         self._log(f"載入嵌入模型: {embed_model_name}")
         self.embed_tokenizer = AutoTokenizer.from_pretrained(embed_model_name)
         self.embed_model = AutoModel.from_pretrained(embed_model_name)
@@ -81,21 +81,67 @@ class TextToSQLSystem:
         # 3. 載入數據集並建立索引
         self.dataset, self.faiss_index = self._load_and_index_dataset()
 
-        # 4. 載入 GGUF 模型
-        self._log("載入 GGUF 模型...")
-        model_path = hf_hub_download(
-            repo_id=GGUF_REPO_ID,
-            filename=GGUF_FILENAME,
-            repo_type="dataset"
-        )
-        self.llm = Llama(
-            model_path=model_path,
-            n_ctx=1024,
-            n_threads=os.cpu_count(),
-            n_batch=512,
-            verbose=False
-        )
+        # 4. 載入 GGUF 模型（添加錯誤處理）
+        self._load_gguf_model()
+
         self._log("✅ 系統初始化完成")
+    def _load_gguf_model(self):
+        """載入 GGUF 模型並處理錯誤"""
+        try:
+            self._log("載入 GGUF 模型...")
+            model_path = hf_hub_download(
+                repo_id=GGUF_REPO_ID,
+                filename=GGUF_FILENAME,
+                repo_type="dataset"
+            )
+            
+            # 檢查文件完整性
+            file_size = os.path.getsize(model_path)
+            expected_size = 986 * 1024 * 1024  # 986MB
+            if file_size != expected_size:
+                self._log(f"⚠️ 文件大小不匹配: {file_size} != {expected_size}", "WARNING")
+                # 重新下載
+                os.remove(model_path)
+                model_path = hf_hub_download(
+                    repo_id=GGUF_REPO_ID,
+                    filename=GGUF_FILENAME,
+                    repo_type="dataset",
+                    force_download=True
+                )
+            
+            # 使用更兼容的參數
+            self.llm = Llama(
+                model_path=model_path,
+                n_ctx=1024,
+                n_threads=max(2, os.cpu_count() - 1),  # 留一個核心給系統
+                n_batch=256,
+                verbose=True,  # 開啟詳細日誌
+                n_gpu_layers=0  # 強制使用CPU
+            )
+            self._log("✅ GGUF 模型載入成功")
+            
+        except Exception as e:
+            self._log(f"❌ GGUF 模型載入失敗: {e}", "ERROR")
+            self._log("嘗試使用備用載入方式...")
+            self._load_gguf_model_fallback(model_path)
+    def _load_gguf_model_fallback(self, model_path):
+        """備用載入方式"""
+        try:
+            # 嘗試不同的參數組合
+            self.llm = Llama(
+                model_path=model_path,
+                n_ctx=512,  # 更小的上下文
+                n_threads=4,
+                n_batch=128,
+                vocab_only=False,
+                use_mmap=True,
+                use_mlock=False,
+                verbose=True
+            )
+            self._log("✅ 備用方式載入成功")
+        except Exception as e:
+            self._log(f"❌ 備用方式也失敗: {e}", "ERROR")
+            self.llm = None
 
     def _log(self, message: str, level: str = "INFO"):
         self.log_history.append(format_log(message, level))