added history

app.py

@@ -1,4 +1,4 @@
-import chromadb
+import chromadb 
 import pandas as pd
 from sentence_transformers import SentenceTransformer
 from langchain.text_splitter import RecursiveCharacterTextSplitter
@@ -8,7 +8,7 @@ from openai import OpenAI
 import numpy as np
 import requests
 import chromadb
-from chromadb import Client 
+from chromadb import Client
 from sentence_transformers import SentenceTransformer, util
 from langchain_community.embeddings import HuggingFaceEmbeddings
 from chromadb import Client
@@ -21,11 +21,8 @@ import requests
 import time
 import tempfile
 
-#HF_TOKEN  = os.getenv("HF_TOKEN")
-
 API_KEY = os.environ.get("OPENROUTER_API_KEY")
 
-
 # Load the Excel file
 df = pd.read_excel("web_documents.xlsx", engine='openpyxl')
 
@@ -38,14 +35,8 @@ collection = client.get_or_create_collection(
     metadata={"hnsw:space": "cosine"}
 )
 
-# Load the embedding model new model 
-#embedding_model = SentenceTransformer('sentence-transformers/multi-qa-mpnet-base-dot-v1')
-#embedding_model = SentenceTransformer("BAAI/bge-m3")
-embedding_model = SentenceTransformer("sentence-transformers/multi-qa-MiniLM-L6-cos-v1")
-
-
-
-
+# Load the embedding model
+embedding_model = SentenceTransformer('sentence-transformers/paraphrase-MiniLM-L6-v2')
 
 # Initialize the text splitter
 text_splitter = RecursiveCharacterTextSplitter(chunk_size=1200, chunk_overlap=150)
@@ -78,21 +69,12 @@ for idx, row in df.iterrows():
 
 # ---------------------- Config ----------------------
 SIMILARITY_THRESHOLD = 0.80
-
-client1 = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=API_KEY)  # remplace par ta clé OpenRouter
-
+client1 = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=API_KEY)  # Replace with your OpenRouter API key
 
 # ---------------------- Models ----------------------
-# High-accuracy model for semantic search
-#semantic_model = SentenceTransformer("sentence-transformers/all-mpnet-base-v2")
-#semantic_model = SentenceTransformer("sentence-transformers/paraphrase-multilingual-mpnet-base-v2")
-semantic_model = SentenceTransformer("sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")
-
+semantic_model = SentenceTransformer("sentence-transformers/all-mpnet-base-v2")
 
-# For ChromaDB
-#embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/multi-qa-mpnet-base-dot-v1")
-
-# ---------------------- Load QA Data ----------------------
+# Load QA Data
 with open("qa.json", "r", encoding="utf-8") as f:
     qa_data = json.load(f)
 
@@ -100,7 +82,7 @@ qa_questions = list(qa_data.keys())
 qa_answers = list(qa_data.values())
 qa_embeddings = semantic_model.encode(qa_questions, convert_to_tensor=True)
 
-# ---------------------- CAG ----------------------
+# ---------------------- History-Aware CAG ----------------------
 def retrieve_from_cag(user_query):
     query_embedding = semantic_model.encode(user_query, convert_to_tensor=True)
     cosine_scores = util.cos_sim(query_embedding, qa_embeddings)[0]
@@ -109,21 +91,19 @@ def retrieve_from_cag(user_query):
 
     print(f"[CAG] Best score: {best_score:.4f} | Closest question: {qa_questions[best_idx]}")
     if best_score >= SIMILARITY_THRESHOLD:
-        return qa_answers[best_idx], best_score
+        return qa_answers[best_idx], best_score  # Only return the answer
     else:
         return None, best_score
 
-# ---------------------- RAG ----------------------
-#client = chromadb.Client()
-#collection = client.get_collection(name="rag_web_db_cosine_full_documents")
-# Assuming you have a persistent Chroma client setup
-#client = PersistentClient("./db_new/db_new")# Replace with the correct path if needed
-#collection = client.get_collection(name="rag_web_db_cosine_full_documents")
-# ---------------------- RAG retrieval ----------------------
-def retrieve_from_rag(user_query):
-    print("Searching in RAG...")
-
-    query_embedding = embedding_model.encode(user_query)
+# ---------------------- History-Aware RAG ----------------------
+def retrieve_from_rag(user_query, chat_history):
+    # Combine the previous chat history with the current query for context
+    history_context = " ".join([f"User: {msg[0]} Bot: {msg[1]}" for msg in chat_history]) + " "
+    full_query = history_context + user_query
+
+    print("Searching in RAG with history context...")
+
+    query_embedding = embedding_model.encode(full_query)
     results = collection.query(query_embeddings=[query_embedding], n_results=3)
 
     if not results or not results.get('documents'):
@@ -151,11 +131,8 @@ Instructions:
 - Use only the provided documents below to answer.
 - If the answer is not in the documents, simply say: "I don't know." / "Je ne sais pas."
 - Cite only the sources you use, indicated at the end of each document like (Source: https://example.com).
-
-
 Documents :
 {context}
-
 Question : {query}
 Answer :
 [/INST]
@@ -171,60 +148,20 @@ Answer :
         print(f"Erreur lors de la génération : {e}")
         return "Erreur lors de la génération."
 
-# ---------------------- Generation function (Huggingface) ----------------------
-def generate_via_huggingface(context, query, max_new_tokens=512, hf_token="your_huggingface_token"):
-    print("\n--- Generating via Huggingface ---")
-    print("Context received:", context)
-
-    prompt = f"""<s>[INST]
-You are a Moodle expert assistant.
-
-Rules:
-- Answer only based on the provided documents.
-- If the answer is not found, reply: "I don't know."
-- Only cite sources mentioned (metadata 'source').
-
-Documents:
-{context}
-
-Question: {query}
-Answer:
-[/INST]
-"""
-
-    API_URL = "https://api-inference.huggingface.co/models/mistralai/Mixtral-8x7B-Instruct-v0.1"
-    headers = {"Authorization": f"Bearer {hf_token}"}
-    payload = {
-        "inputs": prompt,
-        "parameters": {
-            "max_new_tokens": max_new_tokens
-        }
-    }
-
-    response = requests.post(API_URL, headers=headers, json=payload)
-
-    if response.status_code == 200:
-        result = response.json()
-        if isinstance(result, list) and "generated_text" in result[0]:
-            return result[0]["generated_text"].strip()
-        else:
-            return "Error: Unexpected response format."
-    else:
-        return f"Error {response.status_code}: {response.text}"
-
 # ---------------------- Main Chatbot ----------------------
-def chatbot(query):
+def chatbot(query, chat_history):
     print("\n==== New Query ====")
     print("User Query:", query)
 
     # Try to retrieve from CAG (cache)
-    answer, score = retrieve_from_cag(query)
+    answer, score = retrieve_from_cag(query, chat_history)
     if answer:
         print("Answer retrieved from CAG cache.")
+        chat_history.append((query, answer))  # Append the new question-answer pair to history
         return answer
 
     # If not found, retrieve from RAG
-    docs = retrieve_from_rag(query)
+    docs = retrieve_from_rag(query, chat_history)
     if docs:
         context_blocks = []
         for doc in docs:
@@ -241,47 +178,17 @@ def chatbot(query):
 
         context = "\n\n".join(context_blocks)
 
-        # Choose the generation backend (OpenRouter or Huggingface)
+        # Choose the generation backend (OpenRouter)
         response = generate_via_openrouter(context, query)
+        chat_history.append((query, response))  # Append the new question-answer pair to history
         return response
 
     else:
         print("No relevant documents found.")
+        chat_history.append((query, "Je ne sais pas."))
         return "Je ne sais pas."
 
-
 # ---------------------- Gradio App ----------------------
-
-# Define the chatbot response function
-#def ask(user_message, chat_history):
- #   if not user_message:
-  #      return chat_history, chat_history, ""
- #   
-    # Get chatbot response
- #   response = chatbot(user_message)
-
-    # Update chat history
-  #  chat_history.append((user_message, response))
-   # return chat_history, chat_history, ""
-
-# Initialize chat history with a welcome message
-#initial_message = (None, "Hello, how can I help you with Moodle?")
-
-# Build Gradio interface
-#with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    #chat_history = gr.State([initial_message])  # <-- Move inside here!
-
-  #  chatbot_ui = gr.Chatbot(value=[initial_message])
-  #  question = gr.Textbox(placeholder="Ask me anything about Moodle...", show_label=False)
-   # clear_button = gr.Button("Clear")
-
-  #  question.submit(ask, [question, chat_history], [chatbot_ui, chat_history, question])
-   # clear_button.click(lambda: ([initial_message], [initial_message], ""), None, [chatbot_ui, chat_history, question], queue=False)
-
-#demo.queue()
-#demo.launch(share=False)
-# Initialize chat history with a welcome message
-
 def save_chat_to_file(chat_history):
     timestamp = time.strftime("%Y%m%d-%H%M%S")
     filename = f"chat_history_{timestamp}.json"
@@ -294,17 +201,18 @@ def save_chat_to_file(chat_history):
     with open(file_path, "w", encoding="utf-8") as f:
         json.dump(chat_history, f, ensure_ascii=False, indent=2)
 
-    return file_path  # THIS should be only the path, not a tuple!
+    return file_path
 
 def ask(user_message, chat_history):
     if not user_message:
         return chat_history, chat_history, ""
 
-    response = chatbot(user_message)
+    response = chatbot(user_message, chat_history)
     chat_history.append((user_message, response))
 
     return chat_history, chat_history, ""
 
+# Initialize chat history with a welcome message
 initial_message = (None, "Hello, how can I help you with Moodle?")
 
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
@@ -313,11 +221,11 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     chatbot_ui = gr.Chatbot(value=[initial_message])
     question = gr.Textbox(placeholder="Ask me anything about Moodle...", show_label=False)
     clear_button = gr.Button("Clear")
-    save_button = gr.Button("Save Chat")  
+    save_button = gr.Button("Save Chat")
 
     question.submit(ask, [question, chat_history], [chatbot_ui, chat_history, question])
     clear_button.click(lambda: ([initial_message], [initial_message], ""), None, [chatbot_ui, chat_history, question], queue=False)
-    
+
     save_button.click(save_chat_to_file, [chat_history], gr.File(label="Download your chat history"))
 
 demo.queue()