update gemini

rag_core/embedder.py

@@ -6,16 +6,33 @@ from rag_core.utils import log_timed
 
 @log_timed("gửi API tạo embedding")
 def get_embedding(text: str, retries: int = 3, base_timeout: int = 30):
+    api_key = os.getenv("GEMINI_API_KEY")
+    if not api_key:
+        raise ValueError("Thiếu biến môi trường GEMINI_API_KEY.")
+
+    url = f"https://generativelanguage.googleapis.com/v1/models/text-embedding-004:embedContent?key={api_key}"
+
+    payload = {
+        "model": "models/text-embedding-004",
+        "content": {
+            "parts": [
+                { "text": text }
+            ]
+        }
+    }
+
     for i in range(retries):
         try:
             current_timeout = base_timeout * (i + 1)
-            response = requests.post(
-                "https://generativelanguage.googleapis.com/v1/models/text-embedding-004:embedContent?key="+os.getenv("GEMINI_API_KEY"),
-                json={"content": { "parts": [ { "text": text } ] }},
-                timeout=current_timeout
-            )
+            response = requests.post(url, json=payload, timeout=current_timeout)
             response.raise_for_status()
-            return response.json()["embedding"]
+
+            # Kiểm tra phản hồi chứa trường embedding không
+            data = response.json()
+            if "embedding" not in data:
+                raise ValueError(f"Phản hồi không có 'embedding': {data}")
+            return data["embedding"]['values']
+
         except requests.exceptions.RequestException as e:
             logging.warning(f"Lỗi embedding (lần {i+1}/{retries}, timeout={current_timeout}s): {e}")
             if i < retries - 1:
@@ -23,3 +40,4 @@ def get_embedding(text: str, retries: int = 3, base_timeout: int = 30):
             else:
                 raise
 
+