remove flashinfer

generate_summaries_uv.py

@@ -2,7 +2,6 @@
 # requires-python = ">=3.10"
 # dependencies = [
 #     "datasets",
-#     "flashinfer-python",
 #     "hf_transfer",
 #     "huggingface-hub[hf_xet]",
 #     "polars",
@@ -21,7 +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"
+# Note: VLLM will auto-select the best available attention backend
 
 import polars as pl
 from datasets import Dataset, load_dataset