Fix bugs

client.py

@@ -65,7 +65,7 @@ def stream_query(query, user_id=None):
             'query': query,
             'history': []
         }
-        resp = requests.get(url,
+        resp = requests.post(url,
                             headers=headers,
                             data=json.dumps(data),
                             timeout=300,

llm_service/http_client.py

@@ -4,7 +4,7 @@ import json
 import httpx
 import configparser
 import torch
-
+import requests
 import numpy as np
 from loguru import logger
 from BCEmbedding.tools.langchain import BCERerank
@@ -48,9 +48,9 @@ class OpenAPIClient:
                         continue
                     yield output
 
-    async def get_response(self, headers, data):
-        async with httpx.AsyncClient() as client:
-            resp = await client.post(self.url, json=data, headers=headers, timeout=300)
+    def get_response(self, headers, data):
+
+        resp = requests.post(self.url, json=data, headers=headers, timeout=300)
         try:
             result = json.loads(resp.content.decode("utf-8"))
             output = result['choices'][0]['message']['content']

llm_service/worker.py

@@ -137,10 +137,8 @@ class Worker:
 
             if len(chunks) == 0:
                 logger.debug('Response by finetune model')
-                response = self.response_by_finetune(query, history=history)
-                data = json.loads(response.content.decode("utf-8"))
-
-                chunks = [output]
+                response = await self.response_by_finetune(query, history=history)
+                chunks = [response]
             elif use_template:
                 logger.debug('Response by template')
                 response = self.format_rag_result(chunks, references, stream=stream)

rag/retriever.py

@@ -14,7 +14,7 @@ from BCEmbedding.tools.langchain import BCERerank
 from langchain.retrievers import ContextualCompressionRetriever
 from langchain_community.vectorstores.utils import DistanceStrategy
 from requests.exceptions import RequestException
-from elastic_keywords_search import ElasticKeywordsSearch
+from .elastic_keywords_search import ElasticKeywordsSearch
 
 app = FastAPI()