Update app.py

app.py

@@ -1,4 +1,5 @@
 import gradio as gr
+from sentence_transformers import SentenceTransformer
 from huggingface_hub import InferenceClient
 import pandas as pd
 import torch
@@ -9,6 +10,8 @@ For more information on `huggingface_hub` Inference API support, please check th
 """
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
+def get_detailed_instruct(task_description: str, query: str) -> str:
+        return f'Instruct: {task_description}\nQuery: {query}'
 
 def respond(
     message,
@@ -17,37 +20,60 @@ def respond(
     temperature = 0.7,
     top_p = 0.95,
 ):
+    #system role
     messages = [{"role": "system", "content": "You are a moslem bot that always give answer based on quran and hadith!"}]
-    df = pd.read_csv("moslem-bot-reference.csv")
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
 
+    #make a moslem bot
     messages.append({"role": "user", "content": "I want you to answer strictly based on quran and hadith"})
     messages.append({"role": "assistant", "content": "I'd be happy to help! Please go ahead and provide the sentence you'd like me to analyze. Please specify whether you're referencing a particular verse or hadith (Prophetic tradition) from the Quran or Hadith, or if you're asking me to analyze a general statement."})
 
+    #adding references
+    df = pd.read_csv("moslem-bot-reference.csv")
     for index, row in df.iterrows():
         messages.append({"role": "user", "content": row['user']})
         messages.append({"role": "assistant", "content": row['assistant']})
 
-    """selected_dfs = torch.load('selected_dfs.sav', map_location=torch.device('cpu'))
-    for df in selected_dfs:
-        df = df.dropna()
-        n = math.floor(df.shape[0]/10000)
-        print(n)
-        df = df.sample(n)
-        for index, row in df.iterrows():
-            print(index)
-            print(row['Column1.question'])
-            print(row['Column1.answer'])
-            messages.append({"role": "user", "content": row['Column1.question']})
-            messages.append({"role": "assistant", "content": row['Column1.answer']})
-    """
+    #adding more references
+    selected_references = torch.load('selected_references.sav', map_location=torch.device('cpu'))
+    encoded_questions = torch.load('encoded_questions.sav', map_location=torch.device('cpu'))
+    
+    task = 'Given a web search query, retrieve relevant passages that answer the query'
+    queries = [
+        get_detailed_instruct(task, message)
+    ]
+
+    model = SentenceTransformer('intfloat/multilingual-e5-large-instruct')
+    query_embeddings = model.encode(queries, convert_to_tensor=True, normalize_embeddings=True)
+    scores = (query_embeddings @ encoded_questions.T) * 100
+    selected_references['similarity'] = scores.tolist()[0]
+    sorted_references = selected_references.sort_values(by='similarity', ascending=False)
+    sorted_references = sorted_references.head(3)
+    sorted_references = selected_references.sort_values(by='similarity', ascending=True)
+
+    from googletrans import Translator
+    translator = Translator()
+    
+    for index, row in sorted_references.iterrows():
+        print(index)
+        print(row['user'])
+        user = translator.translate(row['user'])
+        print(user)
+        print(row['assistant'])
+        assistant = translator.translate(row['assistant'])
+        print(assistant)
+        messages.append({"role": "user", "content":user })
+        messages.append({"role": "assistant", "content": assistant})
+
+    #history from chat session
+    for val in history:
+        if val[0]:
+            messages.append({"role": "user", "content": val[0]})
+        if val[1]:
+            messages.append({"role": "assistant", "content": val[1]})
+
+    #latest user question
     messages.append({"role": "user", "content": message})
-    #print(messages)
+    print(messages)
 
     response = ""
 
@@ -92,6 +118,5 @@ demo = gr.ChatInterface(
             ],
 )
 
-
 if __name__ == "__main__":
     demo.launch()