Update app.py

app.py

@@ -202,12 +202,12 @@ def get_llama_response(message):
   query_text = message
 
   
-  '''results = db.similarity_search_with_relevance_scores(query_text, k=3)
+  results = db.similarity_search_with_relevance_scores(query_text, k=3)
   if len(results) == 0 or results[0][1] < 0.5:
       print(f"Unable to find matching results.")
 
-  '''
-  #context_text = "\n\n---\n\n".join([doc.page_content for doc, _score in results])
+  
+  context_text = "\n\n---\n\n".join([doc.page_content for doc, _score in results])
   
   template = """
   The following is a conversation between a human an AI. The AI acts exactly like Dwight K Schrute from the TV show The Office.
@@ -225,9 +225,18 @@ def get_llama_response(message):
   prompt = PromptTemplate(input_variables=["history", "input"], template=template+'ss'+ s)
 
   #print(template)
-  chain.prompt=prompt
-  res = chain(query_text)
-  return(res["response"])
+  inputs = tokenizer(query_text, return_tensors="pt").to("cuda")
+
+# Generate text
+  with torch.no_grad():
+   outputs = model.generate(inputs.input_ids, max_length=50)
+
+# Decode the generated text
+  generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
+   
+  #chain.prompt=prompt
+  #res = chain(query_text)
+  return(generated_text)
 
 import gradio as gr