Update app.py

app.py

@@ -460,7 +460,7 @@ def get_response_from_pdf(query, model, selected_docs, num_calls=3, temperature=
         yield "No documents available. Please upload PDF documents to answer questions."
         return
 
-    retriever = database.as_retriever(search_kwargs={"k": 5})
+    retriever = database.as_retriever(search_kwargs={"k": 10})  # Increased k to 10
     logging.info(f"Retrieving relevant documents for query: {query}")
     relevant_docs = retriever.get_relevant_documents(query)
     logging.info(f"Number of relevant documents retrieved: {len(relevant_docs)}")
@@ -476,40 +476,56 @@ def get_response_from_pdf(query, model, selected_docs, num_calls=3, temperature=
 
     for doc in filtered_docs:
         logging.info(f"Document source: {doc.metadata['source']}")
-        logging.info(f"Document content preview: {doc.page_content[:100]}...")  # Log first 100 characters of each document
+        logging.info(f"Document content preview: {doc.page_content[:100]}...")
 
-    context_str = "\n".join([doc.page_content for doc in filtered_docs])
-    logging.info(f"Total context length: {len(context_str)}")
+    # Implement a sliding window approach for context
+    max_context_length = 4000  # Adjust based on your model's capacity
+    context_chunks = []
+    current_chunk = ""
+    for doc in filtered_docs:
+        if len(current_chunk) + len(doc.page_content) > max_context_length:
+            context_chunks.append(current_chunk)
+            current_chunk = doc.page_content
+        else:
+            current_chunk += "\n" + doc.page_content
+    if current_chunk:
+        context_chunks.append(current_chunk)
 
-    if model == "@cf/meta/llama-3.1-8b-instruct":
-        logging.info("Using Cloudflare API")
-        # Use Cloudflare API with the retrieved context
-        for response in get_response_from_cloudflare(prompt="", context=context_str, query=query, num_calls=num_calls, temperature=temperature, search_type="pdf"):
-            yield response
-    else:
-        logging.info("Using Hugging Face API")
-        # Use Hugging Face API
-        prompt = f"""Using the following context from the PDF documents:
+    logging.info(f"Number of context chunks: {len(context_chunks)}")
+
+    for i, context_str in enumerate(context_chunks):
+        logging.info(f"Processing context chunk {i+1}/{len(context_chunks)}")
+        logging.info(f"Context chunk length: {len(context_str)}")
+
+        if model == "@cf/meta/llama-3.1-8b-instruct":
+            logging.info("Using Cloudflare API")
+            for response in get_response_from_cloudflare(prompt="", context=context_str, query=query, num_calls=num_calls, temperature=temperature, search_type="pdf"):
+                yield response
+        else:
+            logging.info("Using Hugging Face API")
+            prompt = f"""Using the following context from the PDF documents:
 {context_str}
 Write a detailed and complete response that answers the following user question: '{query}'"""
         
-        client = InferenceClient(model, token=huggingface_token)
+            client = InferenceClient(model, token=huggingface_token)
         
-        response = ""
-        for i in range(num_calls):
-            logging.info(f"API call {i+1}/{num_calls}")
-            for message in client.chat_completion(
-                messages=[{"role": "user", "content": prompt}],
-                max_tokens=10000,
-                temperature=temperature,
-                stream=True,
-            ):
-                if message.choices and message.choices[0].delta and message.choices[0].delta.content:
-                    chunk = message.choices[0].delta.content
-                    response += chunk
-                    yield response  # Yield partial response
+            response = ""
+            for j in range(num_calls):
+                logging.info(f"API call {j+1}/{num_calls}")
+                for message in client.chat_completion(
+                    messages=[{"role": "user", "content": prompt}],
+                    max_tokens=10000,
+                    temperature=temperature,
+                    stream=True,
+                ):
+                    if message.choices and message.choices[0].delta and message.choices[0].delta.content:
+                        chunk = message.choices[0].delta.content
+                        response += chunk
+                        yield response  # Yield partial response
         
-        logging.info("Finished generating response")
+            logging.info("Finished generating response for this context chunk")
+
+    logging.info("Finished processing all context chunks")
 
 def vote(data: gr.LikeData):
     if data.liked: