Update app.py

app.py

@@ -4,7 +4,8 @@ import os
 from gtts import gTTS
 from deep_translator import GoogleTranslator
 import logging
-from llama_index import VectorStoreIndex, Document
+from llama_index import VectorStoreIndex, Document, SimpleDirectoryReader
+from llama_index.node_parser import SimpleNodeParser
 from llama_index.embeddings import HuggingFaceEmbedding
 from llama_index import ServiceContext
 from llama_index.llms import HuggingFaceLLM
@@ -23,10 +24,13 @@ groq_client = Groq(api_key=os.getenv("GROQ_API_KEY"))
 embed_model = HuggingFaceEmbedding(model_name="sentence-transformers/all-MiniLM-L6-v2")
 
 # Initialize a local LLM for indexing purposes
-local_llm = HuggingFaceLLM(model_name="gpt2", tokenizer_name="gpt2")
+local_llm = HuggingFaceLLM(model_name="gpt2", tokenizer_name="gpt2", context_window=512, max_new_tokens=256)
 
-# Initialize the ServiceContext with the local LLM
-service_context = ServiceContext.from_defaults(llm=local_llm, embed_model=embed_model)
+# Set up node parser for chunking
+node_parser = SimpleNodeParser.from_defaults(chunk_size=256, chunk_overlap=20)
+
+# Initialize the ServiceContext with the local LLM and node parser
+service_context = ServiceContext.from_defaults(llm=local_llm, embed_model=embed_model, node_parser=node_parser)
 
 # Initialize the index
 index = None
@@ -60,11 +64,11 @@ audio_language_dict = {
 def index_text(text: str) -> str:
     global index
     try:
-        document = Document(text=text)
+        documents = [Document(text=text)]
         if index is None:
-            index = VectorStoreIndex.from_documents([document], service_context=service_context)
+            index = VectorStoreIndex.from_documents(documents, service_context=service_context)
         else:
-            index.insert(document)
+            index.insert(documents[0])
         return "Text indexed successfully."
     except Exception as e:
         logging.error(f"Error in indexing: {str(e)}")
@@ -82,6 +86,11 @@ def chat_with_context(question: str, model: str) -> str:
         )
         context = query_engine.query(question).response
         
+        # Truncate context if it's too long
+        max_context_length = 2048  # Adjust as needed
+        if len(context) > max_context_length:
+            context = context[:max_context_length] + "..."
+        
         prompt = f"Context: {context}\n\nQuestion: {question}\n\nAnswer:"
         
         chat_completion = groq_client.chat.completions.create(
@@ -92,12 +101,14 @@ def chat_with_context(question: str, model: str) -> str:
                 }
             ],
             model=model,
+            max_tokens=500  # Limit the response length
         )
         return chat_completion.choices[0].message.content
     except Exception as e:
         logging.error(f"Error in chat: {str(e)}")
         return f"Error in chat: {str(e)}"
 
+
 # Translation function
 def translate_text(text, target_lang_code):
     try: