Update app.py

app.py

@@ -1,5 +1,6 @@
 import langchain
-from langchain.embeddings import SentenceTransformerEmbeddings
+#from langchain.embeddings import SentenceTransformerEmbeddings
+from langchain_community.embeddings import HuggingFaceEmbeddings
 from langchain_community.document_loaders import UnstructuredPDFLoader,UnstructuredWordDocumentLoader
 from langchain.indexes import VectorstoreIndexCreator
 from langchain.vectorstores import FAISS
@@ -19,7 +20,7 @@ import spaces
 #import faiss as FAISS
 
 from groq import Groq
-from sentence_transformers import SentenceTransformer
+#from sentence_transformers import SentenceTransformer
 
 tokenizer = tiktoken.encoding_for_model("gpt-3.5-turbo")
 
@@ -38,11 +39,22 @@ text_splitter = RecursiveCharacterTextSplitter(
     separators=["\n\n", "\n", " ", ""]
 )
 
-embeddings = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")
+#embeddings = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")
 #embeddings = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
+embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2", 
+                                    model_kwargs = {'device': 'cuda'},
+                                    encode_kwargs = {'normalize_embeddings': False})
+
 
 foo = Document(page_content='foo is fou!',metadata={"source":'foo source'})
 
+@spaces.GPU
+def similarity_search(db,query, k=5):
+    return db.similarity_search(query, k)
+
+@spaces.GPU
+def similarity_search_with_score(db,query,k=1):
+    return db.similarity_search_with_score(query,k)
 
 def reset_database(ui_session_id):
   session_id = f"PDFAISS-{ui_session_id}"
@@ -67,7 +79,7 @@ def is_duplicate(split_docs,db):
   print(f"DUPLICATE: Treating: {split_docs[0].metadata['source'].split('/')[-1]}")
   for i in range(min(3,len(split_docs))):
     query = split_docs[i].page_content
-    docs = db.similarity_search_with_score(query,k=1)
+    docs = similarity_search_with_score(db,query,k=1)
     _ , score = docs[0]
     epsilon += score
   print(f"DUPLICATE: epsilon: {epsilon}")
@@ -306,7 +318,7 @@ def ask_gpt(query, ui_session_id, history):
       print(f"SESSION: {session_id} database does not exist")
       return f"SESSION: {session_id} database does not exist","",""
 
-    docs = db.similarity_search(query, k=5)
+    docs = similarity_search(db,query, k=5)
 
     documents = "\n\n*-*-*-*-*-*\n\n".join(f"Content: {doc.page_content}\n" for doc in docs)
     system = f"# Instructions\nTake a deep breath and resonate step by step.\nYou are a helpful standard assistant. Your have only one mission and that consists in answering to the user input based on the **provided documents**. If the answer to the question that is asked by the user isn't contained in the **provided documents**, say so but **don't make up an answer**. I chose you because you can say 'I don't know' so please don't do like the other LLMs and don't define acronyms that aren\'t present in the following **PROVIDED DOCUMENTS** double check if it is present before answering. If some of the information can be useful for the user you can tell him.\nFinish your response by **ONE** follow up question that the provided documents could answer.\n\nThe documents are separated by the string \'*-*-*-*-*-*\'. Do not provide any explanations or details.\n\n# **Provided documents**: {documents}."