Update app.py

app.py

@@ -4,12 +4,12 @@ from PyPDF2 import PdfReader
 from langchain.text_splitter import CharacterTextSplitter
 from langchain_huggingface import HuggingFaceEmbeddings
 from langchain_community.vectorstores import FAISS
-from langchain_community.chat_models import HuggingFacePipeline
+from langchain_huggingface import HuggingFaceHub
 from langchain.memory import ConversationBufferMemory
 from langchain.chains import ConversationalRetrievalChain
-from transformers import pipeline  # Hugging Face pipeline for using T5 model
 import os
 
+
 # Access Hugging Face API token from Streamlit secrets
 hf_token = st.secrets["huggingface"]["HF_TOKEN"]
 
@@ -44,15 +44,21 @@ def get_vectorstore(text_chunks):
 
 # Function to create the conversation chain using T5 from Hugging Face API
 def get_conversation_chain(vectorstore):
-    # Load the T5 model using Hugging Face's pipeline for text generation
-    t5_model = pipeline("text2text-generation", model="t5-small", tokenizer="t5-small", device=0)  # Running on CPU (device=0 for GPU)
-
-    # Use ConversationBufferMemory to track the conversation
-    memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
+    llm = HuggingFaceHub(
+        repo_id="google/t5-large",  # Replace with your Hugging Face model ID
+        temperature=0.7,            # Adjust for creativity
+        max_new_tokens=100,         # Control response length
+        top_p=0.9,                  # Nucleus sampling for better variety
+        top_k=50,                   # Top-k filtering
+        repetition_penalty=1.0      # Default value to prevent repetitive outputs
+    )
+    memory = ConversationBufferMemory(
+        memory_key="chat_history", return_messages=True
+    )
     
-    # Create a conversation chain using the T5 model
+    # Create a conversation chain using the correct LLM (llm, not t5_model)
     conversation_chain = ConversationalRetrievalChain.from_llm(
-        llm=t5_model,
+        llm=llm,  # Pass 'llm' instead of 't5_model'
         retriever=vectorstore.as_retriever(),
         memory=memory,
     )