using semantic_search

app.py

@@ -7,6 +7,7 @@ from langchain_openai import ChatOpenAI
 from langchain_core.prompts import PromptTemplate
 from langchain_core.messages import HumanMessage, SystemMessage
 from sentence_transformers import util
+from torch import tensor
 
 
 
@@ -39,9 +40,9 @@ def get_chat_api(api_key:str):
 
 def get_results(embeddings_model,input,df,num_of_results) -> pd.DataFrame:
     embeddings = embeddings_model.embed_query('query: '+ input)
-    df['similarity'] = df['embeddings'].apply(lambda x: util.dot_score(x,embeddings))
-    results = df.sort_values(by='similarity', ascending=False)
-    return results.head(num_of_results)
+    hits = util.semantic_search(tensor(embeddings), tensor(df['embeddings'].tolist()), top_k=num_of_results)
+    hit_list = [hit['corpus_id'] for hit in hits[0]]
+    return df.iloc[hit_list]
 
 def get_llm_results(query,chat,results):