A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 6, 08:15 UTC
[MRV2] Avoid recompilation of _gather_block_tables_kernel (#37645)
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Woosuk Kwon committed
Mar 20, 2026 at 10:31 UTC
e80cfe575d52232b558a053f9c6c12ebd5b6b081
1 file changed
+1
-1
vllm/v1/worker/gpu/block_table.py
+1
-1
@@ -169,7 +169,7 @@ class BlockTables:
169
return self.slot_mappings[:, :num_tokens]
170
171
172
-@triton.jit
172
+@triton.jit(do_not_specialize=["num_reqs"])
173
def _gather_block_tables_kernel(
174
batch_idx_to_req_idx, # [batch_size]
175
src_block_table_ptrs, # [num_kv_cache_groups]