don't set flashinfer

generate_summaries_uv.py

@@ -20,7 +20,7 @@ from typing import Optional
 
 # Set environment variables to speed up model loading
 os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
-os.environ["VLLM_ATTENTION_BACKEND"] = "FLASHINFER"
+# os.environ["VLLM_ATTENTION_BACKEND"] = "FLASHINFER"
 
 import polars as pl
 from datasets import Dataset, load_dataset
@@ -231,6 +231,7 @@ if __name__ == "__main__":
         print("export HOME=/tmp && \\")
         print("export USER=dummy && \\")
         print("export TORCHINDUCTOR_CACHE_DIR=/tmp/torch-inductor && \\")
+        print("export UV_TORCH_BACKEND=auto && \\")
         print("uv run generate_summaries_uv.py \\")
         print("  davanstrien/Smol-Hub-tldr \\")
         print("  librarian-bots/dataset_cards_with_metadata \\")