[WIP] remove num_prefetch_blocks varible.

2025-12-24 18:22:26 +08:00
parent b264de903d
commit 782437c486
10 changed files with 465 additions and 18 deletions
--- a/bench_offload.py
+++ b/bench_offload.py
@@ -109,7 +109,6 @@ def main():
        max_num_batched_tokens=max_len,
        enable_cpu_offload=True,
        num_gpu_blocks=8,  # Small GPU buffer for offload testing
-        num_prefetch_blocks=4,
    )

    if not args.no_sparse: