fix: run in hf

app.py

@@ -27,9 +27,10 @@ def load_models():
     )
     model = RagSequenceForGeneration.from_pretrained(
         "facebook/rag-sequence-nq",
-        retriever=retriever,
-        device_map="auto"
+        retriever=retriever
     )
+    # Move to CPU (since we're in a CPU environment)
+    model = model.cpu()
     return tokenizer, retriever, model
 
 @st.cache_data  # Cache dataset on disk

requirements.txt

@@ -4,4 +4,5 @@ datasets
 sentence-transformers
 faiss-cpu
 arxiv
-torch
+torch
+accelerate>=0.26.0