住所処理の埋め込み機能を改善し、リクエストを2048件ずつ分割して処理するように変更。全てのレスポンスを統合して返すように修正しました。

app.py

@@ -221,23 +221,28 @@ class InferenceEndpointError(Exception):
 
 def embed_via_multilingual_e5_large(query_addresses):
     headers = {
-        "Accept" : "application/json",
+        "Accept": "application/json",
         "Authorization": f"Bearer {HUGGING_FACE_TOKEN}",
-        "Content-Type": "application/json" 
+        "Content-Type": "application/json"
     }
 
-    response = requests.post(EMBEDDING_MODEL_ENDPOINT, headers=headers, json={"inputs": query_addresses})
-    response_json = response.json()
-    
-    if 'error' in response_json:
-        if response_json['error'] == 'Bad Request: Invalid state':
-            raise InferenceEndpointError(InferenceEndpointErrorCode.INVALID_STATE, "Bad Request: Invalid state")
-        elif response_json['error'] == '503 Service Unavailable':
-            raise InferenceEndpointError(InferenceEndpointErrorCode.SERVICE_UNAVAILABLE, "Service Unavailable")
-        else:
-            raise InferenceEndpointError(InferenceEndpointErrorCode.UNKNOWN_ERROR, response_json['error'])
-    
-    return response_json
+    all_responses = []
+    for i in range(0, len(query_addresses), 2048):
+        chunk = query_addresses[i:i + 2048]
+        response = requests.post(EMBEDDING_MODEL_ENDPOINT, headers=headers, json={"inputs": chunk})
+        response_json = response.json()
+
+        if 'error' in response_json:
+            if response_json['error'] == 'Bad Request: Invalid state':
+                raise InferenceEndpointError(InferenceEndpointErrorCode.INVALID_STATE, "Bad Request: Invalid state")
+            elif response_json['error'] == '503 Service Unavailable':
+                raise InferenceEndpointError(InferenceEndpointErrorCode.SERVICE_UNAVAILABLE, "Service Unavailable")
+            else:
+                raise InferenceEndpointError(InferenceEndpointErrorCode.UNKNOWN_ERROR, response_json['error'])
+
+        all_responses.extend(response_json)
+
+    return all_responses
 
 def search_via_milvus(query_vector, top_k, collection_name, thresh=0.0):
     search_params = {"metric_type": "COSINE", "params": {"nprobe": 10}}  # MiniLM系はCOSINE推奨