Update src/RAGSample.py

src/RAGSample.py

@@ -18,6 +18,10 @@ import pandas as pd
 from typing import Optional, List
 import re
 
+# OPTION 1: Use Hugging Face Pipeline (Recommended for HF Spaces)
+from transformers import pipeline
+from langchain.llms import HuggingFacePipeline
+
 # Disable ChromaDB telemetry to avoid the error
 os.environ["ANONYMIZED_TELEMETRY"] = "False"
 os.environ["CHROMA_SERVER_HOST"] = "localhost"
@@ -357,16 +361,12 @@ Answer:
 """,
         input_variables=["question", "documents"],
     )
-
-    # OPTION 1: Use Hugging Face Pipeline (Recommended for HF Spaces)
-    from transformers import pipeline
-    from langchain.llms import HuggingFacePipeline
     
     # Initialize a local Hugging Face model
     hf_pipeline = pipeline(
         "text-generation",
-        model="microsoft/DialoGPT-medium",  # Good for Q&A tasks
-        tokenizer="microsoft/DialoGPT-medium",
+        model="deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",  # Good for Q&A tasks
+        tokenizer="deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
         max_length=512,
         temperature=0.1,
         device=0 if torch.cuda.is_available() else -1,