Update app.py

app.py

@@ -42,12 +42,30 @@ embeddings = HuggingFaceEmbeddings(
     encode_kwargs=encode_kwargs
 )
 
+index = faiss.load_index("resourse//embeddings_ngap.faiss")
+
+def get_text_embedding(text):
+
+    return embeddings.embed_query(text)
+
 def generater(message, history, temperature, top_p, top_k):
     prompt = "<s>"
     for user_message, assistant_message in history:
         prompt += model.config["promptTemplate"].format(user_message)
-        prompt += assistant_message + "</s>"
+
+        question = prompt
+        question_embeddings = np.array([get_text_embedding(prompt)])
+        D, I = index.search(question_embeddings, k=2) # distance, index
+        retrieved_chunk = [chunks[i] for i in I.tolist()[0]]
+
+        
+        
+        prompt += assistant_message + " Contexte:" + retrieved_chunk + "</s>"
     prompt += model.config["promptTemplate"].format(message)
+
+    
+    
+    
     outputs = []    
     for token in model.generate(prompt=prompt, temp=temperature, top_k = top_k, top_p = top_p, max_tokens = max_new_tokens, streaming=True):
         outputs.append(token)