modify the pipeline

app.py

@@ -121,12 +121,6 @@ def get_retrieval_qa_chain(text_file, hf_model):
     if text_file != default_text_file or default_text_file is None:
         if text_file is not None and os.path.exists(text_file):
             retriever, vectorstore = prepare_vector_store_retriever(text_file)
-        else:
-            # Create a dummy retriever if no file is available
-            dummy_doc = Document(page_content="No document loaded. Please upload a file to get started.")
-            dummy_vectorstore = FAISS.from_documents([dummy_doc], embeddings)
-            retriever = VectorStoreRetriever(vectorstore=dummy_vectorstore, search_kwargs={"k": 1})
-            vectorstore = dummy_vectorstore
 
     chain = RetrievalQA.from_chain_type(
         llm=hf_model,
@@ -146,15 +140,12 @@ def generate(question, answer, text_file, max_new_tokens):
         # Create pipeline without streamer first to test
         phi2_pipeline = pipeline(
             "text-generation",
-            tokenizer=tokenizer,
             model=model,
+            tokenizer=tokenizer,
             max_new_tokens=max_new_tokens,
+            do_sample=False,               # ← greedy
             pad_token_id=tokenizer.eos_token_id,
             eos_token_id=tokenizer.eos_token_id,
-            do_sample=True,
-            temperature=0.7,
-            top_p=0.9,
-            repetition_penalty=1.1,
         )
 
         hf_model = HuggingFacePipeline(pipeline=phi2_pipeline)