Updated app.py

app.py

@@ -1,8 +1,3 @@
-import os
-os.system("langchain upgrade-imports")
-os.system("pip install accelerate")
-os.system("pip install -i https://pypi.org/simple/ bitsandbytes")
-
 import gradio as gr
 # import fitz  # PyMuPDF for extracting text from PDFs
 from langchain.embeddings import HuggingFaceEmbeddings
@@ -61,12 +56,14 @@ query_pipeline = transformers.pipeline(
     return_full_text=True,
     torch_dtype=torch.float16,
     device_map=device,
-    temperature=0.7,
+    do_sample=True,  # Enable sampling
+    temperature=0.7,  # Keep if sampling is used
     top_p=0.9,
     top_k=50,
     max_new_tokens=256
 )
 
+
 llm = HuggingFacePipeline(pipeline=query_pipeline)
 
 books_db_client_retriever = RetrievalQA.from_chain_type(
@@ -91,7 +88,9 @@ def test_rag(query):
     return corrected_text_books
 
 # Define the Gradio interface
-def chat(query, history=[]):
+def chat(query, history=None):
+    if history is None:
+        history = []
     answer = test_rag(query)
     history.append((query, answer))
     return history, history