updata

app.py

@@ -3,7 +3,7 @@ import json
 import gradio as gr
 from datetime import datetime
 
-invoke_url = "https://0tjlmoahsh.execute-api.us-west-2.amazonaws.com/prod"
+invoke_url = "https://zhw65mrud3.execute-api.us-west-2.amazonaws.com/prod"
 api = invoke_url + '/langchain_processor_qa?query='
 
 # chinese_index = "smart_search_qa_test_0614_wuyue_2"
@@ -82,20 +82,15 @@ def get_answer(question,session_id,language,prompt,search_engine,index,top_k,tem
         url += '&language=english'
         url += ('&embedding_endpoint_name=pytorch-inference-all-minilm-l6-v2')
         url += ('&llm_embedding_name=pytorch-inference-vicuna-p3-2x')
-    elif language == "chinese-llm-v1":
+    elif language == "chinese":
         url += '&language=chinese'
         url += ('&embedding_endpoint_name=huggingface-inference-text2vec-base-chinese-v1')
-        url += ('&llm_embedding_name=pytorch-inference-chatglm-v1')
+        url += ('&llm_embedding_name=pytorch-inference-chatglm2-g5-2x')
      
-    elif language == "chinese-llm-v2":
-        url += '&language=chinese'
+    elif language == "chinese-tc":
+        url += '&language=chinese-tc'
         url += ('&embedding_endpoint_name=huggingface-inference-text2vec-base-chinese-v1')
-        url += ('&llm_embedding_name=pytorch-inference-chatglm2-g5-2x')   
-
-        # if llm_instance == 'p3-8x':
-        #     url += ('&llm_embedding_name=pytorch-inference-chatglm-v1-p3-8x')
-        # elif llm_instance == 'g4dn-8x':
-        #     url += ('&llm_embedding_name=pytorch-inference-chatglm-v1-8x')
+        url += ('&llm_embedding_name=pytorch-inference-chatglm2-g5-2x')
     
     if len(session_id) > 0:
         url += ('&session_id='+session_id)
@@ -240,7 +235,7 @@ with demo:
                     session_id_textbox = gr.Textbox(label="Session ID")
                     qa_button = gr.Button("Summit")
 
-                    qa_language_radio = gr.Radio(["chinese-llm-v1","chinese-llm-v2", "english"],value="chinese-llm-v1",label="Language")
+                    qa_language_radio = gr.Radio(["chinese","chinese-tc", "english"],value="chinese",label="Language")
                     # qa_llm_radio = gr.Radio(["p3-8x", "g4dn-8x"],value="p3-8x",label="Chinese llm instance")
                     qa_prompt_textbox = gr.Textbox(label="Prompt( must include {context} and {question} )",placeholder=chinese_prompt,lines=2)
                     qa_search_engine_radio = gr.Radio(["OpenSearch","Kendra"],value="OpenSearch",label="Search engine")