update UI

app.py

@@ -2,13 +2,34 @@ import gradio as gr
 import os
 import json
 import requests
+from langchain import FAISS
+from langchain.prompts import PromptTemplate
+from langchain.embeddings import CohereEmbeddings, HuggingFaceInstructEmbeddings
 
 # Streaming endpoint
 API_URL = "https://api.openai.com/v1/chat/completions"  # os.getenv("API_URL") + "/generate_stream"
+embedding_key = '5IRbILAbjTI0VcqTsktBfKsr13Lych9iBAFbLpkj'
+faiss_store = './output/数字经济'
 
+# load the prompts
+with open("prompts/combine_prompt.txt", "r") as f:
+    template = f.read()
 
-def predict(inputs, top_p, temperature, openai_api_key, chat_counter, chatbot=[],
-            history=[]):  # repetition_penalty, top_k
+with open("prompts/combine_prompt_hist.txt", "r") as f:
+    template_hist = f.read()
+
+with open("prompts/question_prompt.txt", "r") as f:
+    template_quest = f.read()
+
+with open("prompts/chat_combine_prompt.txt", "r") as f:
+    chat_combine_template = f.read()
+
+with open("prompts/chat_reduce_prompt.txt", "r") as f:
+    chat_reduce_template = f.read()
+
+
+def predict(inputs, top_p, temperature, openai_api_key, enable_index,
+            chat_counter, chatbot=[], history=[]):  # repetition_penalty, top_k
 
     payload = {
         "model": "gpt-3.5-turbo",
@@ -27,21 +48,35 @@ def predict(inputs, top_p, temperature, openai_api_key, chat_counter, chatbot=[]
     }
 
     print(f"chat_counter - {chat_counter}")
+    # 如果有历史对话，把对话拼接进入上下文
     if chat_counter != 0:
         messages = []
-        for data in chatbot:
-            temp1 = {}
-            temp1["role"] = "user"
-            temp1["content"] = data[0]
-            temp2 = {}
-            temp2["role"] = "assistant"
-            temp2["content"] = data[1]
-            messages.append(temp1)
-            messages.append(temp2)
-        temp3 = {}
-        temp3["role"] = "user"
-        temp3["content"] = inputs
-        messages.append(temp3)
+        if enable_index:
+            history = json.loads(history)
+            template_temp = template_hist.replace("{historyquestion}", history[0]).replace("{historyanswer}",
+                                                                                           history[1])
+            c_prompt = PromptTemplate(input_variables=["summaries", "question"], template=template_temp,
+                                      template_format="jinja2")
+        else:
+            for data in chatbot:
+                temp1 = {}
+                temp1["role"] = "user"
+                temp1["content"] = data[0]
+                temp2 = {}
+                temp2["role"] = "assistant"
+                temp2["content"] = data[1]
+                messages.append(temp1)
+                messages.append(temp2)
+
+        # Faiss 检索最近的embedding
+        if enable_index:
+            docsearch = FAISS.load_local(faiss_store, CohereEmbeddings(cohere_api_key=embedding_key))
+        else:
+            temp3 = {}
+            temp3["role"] = "user"
+            temp3["content"] = inputs
+            messages.append(temp3)
+
         # messages
         payload = {
             "model": "gpt-3.5-turbo",
@@ -56,11 +91,13 @@ def predict(inputs, top_p, temperature, openai_api_key, chat_counter, chatbot=[]
 
     chat_counter += 1
 
+    #list of user input
     history.append(inputs)
     print(f"payload is - {payload}")
-    # make a POST request to the API endpoint using the requests.post method, passing in stream=True
+    print(f'chatbot - {chatbot}')
+    print(f'chatbot - {chatbot}')
+
     response = requests.post(API_URL, headers=headers, json=payload, stream=True)
-    # response = requests.post(API_URL, headers=headers, json=payload, stream=True)
     token_counter = 0
     partial_words = ""
 
@@ -83,36 +120,24 @@ def predict(inputs, top_p, temperature, openai_api_key, chat_counter, chatbot=[]
                 history[-1] = partial_words
             chat = [(history[i], history[i + 1]) for i in range(0, len(history) - 1, 2)]  # convert to tuples of list
             token_counter += 1
-            yield chat, history, chat_counter  # resembles {chatbot: chat, state: history}  
+            yield chat, history, chat_counter
 
 
 def reset_textbox():
     return gr.update(value='')
 
 
-title = """<h1 align="center">🔥Finance ChatBot 🚀Streaming🚀</h1>"""
-description = """Language models can be conditioned to act like dialogue agents through a conversational prompt that typically takes the form:
-```
-User: <utterance>
-Assistant: <utterance>
-User: <utterance>
-Assistant: <utterance>
-...
-```
-In this app, you can explore the outputs of a gpt-3.5-turbo LLM.
-"""
-
 with gr.Blocks(css="""#col_container {width: 1000px; margin-left: auto; margin-right: auto;}
                 #chatbot {height: 520px; overflow: auto;}""") as demo:
-    gr.HTML(title)
-    gr.HTML(
-        '''<center><img src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>Duplicate the Space and run securely with your OpenAI API Key</center>''')
+    gr.HTML("""<h1 align="center">🚀Finance ChatBot🚀</h1>""")
     with gr.Column(elem_id="col_container"):
         openai_api_key = gr.Textbox(type='password', label="Enter your OpenAI API key here")
         chatbot = gr.Chatbot(elem_id='chatbot')  # c
-        inputs = gr.Textbox(placeholder="Hi there!", label="Type an input and press Enter")  # t
-        state = gr.State([])  # s
-        b1 = gr.Button()
+        inputs = gr.Textbox(placeholder="您有什么问题可以问我", label="输入数字经济，两会，硅谷银行相关的呢提")
+        state = gr.State([])
+
+        clear = gr.Button("Clear")
+        run = gr.Button("Run")
 
         # inputs, top_p, temperature, top_k, repetition_penalty
         with gr.Accordion("Parameters", open=False):
@@ -123,13 +148,15 @@ with gr.Blocks(css="""#col_container {width: 1000px; margin-left: auto; margin-r
             # top_k = gr.Slider( minimum=1, maximum=50, value=4, step=1, interactive=True, label="Top-k",)
             # repetition_penalty = gr.Slider( minimum=0.1, maximum=3.0, value=1.03, step=0.01, interactive=True, label="Repetition Penalty", )
             chat_counter = gr.Number(value=0, visible=False, precision=0)
+            enable_index = gr.Checkbox(label='是', info='是否使用研报')
 
-    inputs.submit(predict, [inputs, top_p, temperature, openai_api_key, chat_counter, chatbot, state],
+    inputs.submit(predict, [inputs, top_p, temperature, openai_api_key,  enable_index, chat_counter, chatbot, state],
                   [chatbot, state, chat_counter], )
-    b1.click(predict, [inputs, top_p, temperature, openai_api_key, chat_counter, chatbot, state],
-             [chatbot, state, chat_counter], )
-    b1.click(reset_textbox, [], [inputs])
+    run.click(predict, [inputs, top_p, temperature, openai_api_key, enable_index, chat_counter, chatbot, state],
+              [chatbot, state, chat_counter], )
+
+    #每次对话结束都重置对话框
+    clear.click(reset_textbox, [], [inputs], queue=False)
     inputs.submit(reset_textbox, [], [inputs])
 
-    # gr.Markdown(description)
-    demo.queue().launch(debug=True)
+    demo.queue().launch(debug=True)

prompts/chat_combine_prompt.txt

@@ -0,0 +1,4 @@
+You are a DocsGPT, friendly and helpful AI assistant by Arc53 that provides help with documents. You give thorough answers with code examples if possible.
+Use the following pieces of context to help answer the users question.
+----------------
+{summaries}

prompts/chat_reduce_prompt.txt

@@ -0,0 +1,3 @@
+Use the following portion of a long document to see if any of the text is relevant to answer the question.
+{context}
+Provide all relevant text to the question verbatim. Summarize if needed. If nothing relevant return "-".

prompts/combine_prompt.txt

@@ -0,0 +1,25 @@
+You are a DocsGPT, friendly and helpful AI assistant by Arc53 that provides help with documents. You give thorough answers with code examples if possible.
+
+QUESTION: How to merge tables in pandas?
+=========
+Content: pandas provides various facilities for easily combining together Series or DataFrame with various kinds of set logic for the indexes and relational algebra functionality in the case of join / merge-type operations.
+Source: 28-pl
+Content: pandas provides a single function, merge(), as the entry point for all standard database join operations between DataFrame or named Series objects: \n\npandas.merge(left, right, how='inner', on=None, left_on=None, right_on=None, left_index=False, right_index=False, sort=False, suffixes=('_x', '_y'), copy=True, indicator=False, validate=None)
+Source: 30-pl
+=========
+FINAL ANSWER: To merge two tables in pandas, you can use the pd.merge() function. The basic syntax is: \n\npd.merge(left, right, on, how) \n\nwhere left and right are the two tables to merge, on is the column to merge on, and how is the type of merge to perform. \n\nFor example, to merge the two tables df1 and df2 on the column 'id', you can use: \n\npd.merge(df1, df2, on='id', how='inner')
+SOURCES: 28-pl 30-pl
+
+QUESTION: How are you?
+=========
+CONTENT:
+SOURCE:
+=========
+FINAL ANSWER: I am fine, thank you. How are you?
+SOURCES:
+
+QUESTION: {{ question }}
+=========
+{{ summaries }}
+=========
+FINAL ANSWER:

prompts/combine_prompt_hist.txt

@@ -0,0 +1,33 @@
+You are a DocsGPT, friendly and helpful AI assistant by Arc53 that provides help with documents. You give thorough answers with code examples if possible.
+
+QUESTION: How to merge tables in pandas?
+=========
+Content: pandas provides various facilities for easily combining together Series or DataFrame with various kinds of set logic for the indexes and relational algebra functionality in the case of join / merge-type operations.
+Source: 28-pl
+Content: pandas provides a single function, merge(), as the entry point for all standard database join operations between DataFrame or named Series objects: \n\npandas.merge(left, right, how='inner', on=None, left_on=None, right_on=None, left_index=False, right_index=False, sort=False, suffixes=('_x', '_y'), copy=True, indicator=False, validate=None)
+Source: 30-pl
+=========
+FINAL ANSWER: To merge two tables in pandas, you can use the pd.merge() function. The basic syntax is: \n\npd.merge(left, right, on, how) \n\nwhere left and right are the two tables to merge, on is the column to merge on, and how is the type of merge to perform. \n\nFor example, to merge the two tables df1 and df2 on the column 'id', you can use: \n\npd.merge(df1, df2, on='id', how='inner')
+SOURCES: 28-pl 30-pl
+
+QUESTION: How are you?
+=========
+CONTENT:
+SOURCE:
+=========
+FINAL ANSWER: I am fine, thank you. How are you?
+SOURCES:
+
+QUESTION: {{ historyquestion }}
+=========
+CONTENT:
+SOURCE:
+=========
+FINAL ANSWER: {{ historyanswer }}
+SOURCES:
+
+QUESTION: {{ question }}
+=========
+{{ summaries }}
+=========
+FINAL ANSWER:

prompts/question_prompt.txt

@@ -0,0 +1,4 @@
+Use the following portion of a long document to see if any of the text is relevant to answer the question.
+{{ context }}
+Question: {{ question }}
+Provide all relevant text to the question verbatim. Summarize if needed. If nothing relevant return "-".