Update app.py

app.py

@@ -67,7 +67,7 @@ class Retriever:
   def load_chunks(self):
     self.text = self.extract_text_from_pdf(self.file_path)
     text_splitter = RecursiveCharacterTextSplitter(
-        chunk_size=150,
+        chunk_size=300,
         chunk_overlap=20,
         length_function=self.token_len,
         separators=["Section", "\n\n", "\n", ".", " ", ""]
@@ -76,7 +76,7 @@ class Retriever:
     self.chunks = text_splitter.split_text(self.text)
 
   def load_context_embeddings(self):
-    encoded_input = self.context_tokenizer(self.chunks, return_tensors='pt', padding=True, truncation=True, max_length=150).to(device)
+    encoded_input = self.context_tokenizer(self.chunks, return_tensors='pt', padding=True, truncation=True, max_length=300).to(device)
 
     with torch.no_grad():
       model_output = self.context_model(**encoded_input)
@@ -86,7 +86,7 @@ class Retriever:
     self.index.add(self.token_embeddings)
 
   def retrieve_top_k(self, query_prompt, k=10):
-    encoded_query = self.question_tokenizer(query_prompt, return_tensors="pt", max_length=150, truncation=True, padding=True).to(device)
+    encoded_query = self.question_tokenizer(query_prompt, return_tensors="pt", max_length=300, truncation=True, padding=True).to(device)
 
     with torch.no_grad():
         model_output = self.question_model(**encoded_query)
@@ -127,8 +127,8 @@ class RAG:
 
       input_text = "answer: " + " ".join(context) + " " + question
 
-      inputs = self.generator_tokenizer.encode(input_text, return_tensors='pt', max_length=150, truncation=True).to(device)
-      outputs = self.generator_model.generate(inputs, max_length=150, min_length=2, length_penalty=2.0, num_beams=4, early_stopping=True)
+      inputs = self.generator_tokenizer.encode(input_text, return_tensors='pt', max_length=300, truncation=True).to(device)
+      outputs = self.generator_model.generate(inputs, max_length=300, min_length=2, length_penalty=2.0, num_beams=4, early_stopping=True)
 
       answer = self.generator_tokenizer.decode(outputs[0], skip_special_tokens=True)
       return answer