Update app.py

app.py

@@ -41,9 +41,8 @@ retriever = AutoModelForSeq2SeqLM.from_pretrained(retriever_model_name)
 retriever_tokenizer = AutoTokenizer.from_pretrained(retriever_model_name)
 
 # Initialize FAISS index using LangChain
-embedding_dimension = embedding_model.get_sentence_embedding_dimension()
 hf_embeddings = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L6-v2')
-faiss_index = FAISS(embedding_function=hf_embeddings, dimension=embedding_dimension)
+faiss_index = FAISS(embedding_function=hf_embeddings)
 
 # Load or create FAISS index
 index_path = "faiss_index.pkl"
@@ -74,7 +73,7 @@ def upload_files(files):
 
             # Encode sentences and add to FAISS index
             embeddings = embedding_model.encode(sentences)
-            faiss_index.add_texts(sentences, embeddings)
+            faiss_index.add_texts(sentences)
 
         # Save the updated index
         with open(index_path, "wb") as f:
@@ -96,10 +95,8 @@ def process_and_query(state, files, question):
         question_embedding = embedding_model.encode([question])
 
         # Search the FAISS index for similar passages
-        distances, retrieved_ids = faiss_index.similarity_search_with_score(question_embedding, k=5)  # Retrieve top 5 passages
-
-        # Get the retrieved passages from the document text
-        retrieved_passages = [state["processed_text"].split("\n")[i] for i in retrieved_ids.flatten()]
+        retrieved_results = faiss_index.similarity_search(question, k=5)  # Retrieve top 5 passages
+        retrieved_passages = [result['text'] for result in retrieved_results]
 
         # Use generator model to generate response based on question and retrieved passages
         combined_input = question + " ".join(retrieved_passages)
@@ -132,3 +129,5 @@ with gr.Blocks() as demo:
         query_button.click(fn=process_and_query, inputs=[query], outputs=query_output)
 
 demo.launch()
+
+