Update app.py

app.py

@@ -34,7 +34,7 @@ def extract_text_from_pdf(pdf_file):
         pages.append(f"Error reading PDF: {str(e)}")
     return pages
 
-def chunk_text(text, chunk_size=500):
+def chunk_text(text, chunk_size=1500):
     """Split text into chunks of approximately chunk_size characters."""
     words = text.split()
     chunks = []
@@ -95,7 +95,7 @@ def rag_pipeline(question, pdf_files):
 
     # Retrieve top 3 chunks using cosine similarity
     cos_scores = util.pytorch_cos_sim(query_embedding, doc_embeddings)[0]
-    top_results = torch.topk(cos_scores, k=min(3, len(documents)))
+    top_results = torch.topk(cos_scores, k=min(5, len(documents)))
     retrieved_context = ""
     for score, idx in zip(top_results.values, top_results.indices):
         retrieved_context += f"- {documents[idx]} (score: {score:.2f})\n"
@@ -120,7 +120,7 @@ def rag_pipeline(question, pdf_files):
 
     # Generate answer with more tokens
     inputs = gen_tokenizer(prompt, return_tensors="pt")
-    outputs = gen_model.generate(**inputs, max_new_tokens=150, num_beams=2)
+    outputs = gen_model.generate(**inputs, max_new_tokens=1500, num_beams=2)
     answer = gen_tokenizer.decode(outputs[0], skip_special_tokens=True)
 
     # Log processing time