Update app.py

app.py

@@ -1,7 +1,9 @@
 import os
 from threading import Thread
 from typing import Iterator
-
+from haystack import Document
+from haystack import Pipeline
+from haystack.components.embedders import SentenceTransformersTextEmbedder
 import gradio as gr
 import spaces
 import torch
@@ -22,6 +24,52 @@ if torch.cuda.is_available():
     model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_4bit=True)
     tokenizer = AutoTokenizer.from_pretrained(model_id)
     tokenizer.use_default_system_prompt = False
+    query_pipeline1 = Pipeline()
+    query_pipeline1.add_component(
+        "text_embedder",
+        SentenceTransformersTextEmbedder(
+            model="intfloat/multilingual-e5-large", 
+            prefix="query:",
+        ))
+
+    
+
+
+key='4vNfIDO8PmFwCloxA40y2b4PSHm62vmcuPoM'
+url = "https://mmchpi0yssanukk5t3ofta.c0.europe-west3.gcp.weaviate.cloud"
+# instanziamento client weaviate
+client = weaviate.Client(
+    url = url,  
+    auth_client_secret=weaviate.auth.AuthApiKey(api_key=key),
+    #embedded_options=weaviate.embedded.EmbeddedOptions(),
+)
+
+
+
+
+
+def vettorizz_query(query):
+    vector_query=query_pipeline1.run({   "text_embedder": {"text": query},
+    })['text_embedder']['embedding']
+    return query, vector_query
+
+def richiamo_materiali(query, alpha, N_items):
+    text_query, vett_query = vettorizz_query(query)
+    try:
+        materiali = client.query.get("Default", ["content"]).with_hybrid(
+            query=text_query,
+            vector=vett_query,
+            alpha=alpha,
+            fusion_type=HybridFusion.RELATIVE_SCORE,
+        ).with_additional(["score"]).with_limit(N_items).do()
+        
+        mat = [{'score':i['_additional']['score'],'content':i['content']} for i in materiali['data']['Get']['Default']]
+    except:
+        mat =[{'score':0, 'content':'NESSUN MATERIALE FORNITO'}]
+    
+    return mat
+
+
 
 
 @spaces.GPU
@@ -44,7 +92,19 @@ def generate(
                          Firmati alla fine di ogni risposta '-AvvoChat'.'''})
     for user, assistant in chat_history:
         conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
-    conversation.append({"role": "user", "content": message})
+    materiali = richiamo_materiali(message, alpha=1.0, n_items=5)
+    documenti = ''
+    for idx, d in enumerate(materiali):
+        if idx<len(materiali)-1:
+            documenti += f"{d['content']}; "
+        else:
+            documenti += f"{d['content']}. "
+    text = f'''Basandoti sulle tue conoscenze e usando le informazioni contenute che ti fornisco di seguito.
+    CONTESTO:
+    {documenti}
+    Rispondi in modo esaustivo, evitando inutili giri di parole o ripetizioni, alla seguente domanda.
+    {message}'''
+    conversation.append({"role": "user", "content": text})
 
     input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
     if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH: