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 15, 01:00 UTC
[BugFix] Fix full cuda graph slot_mapping (#21228)
Signed-off-by: fhl2000 <63384265+fhl2000@users.noreply.github.com>
fhl2000 committed
Jul 20, 2025 at 05:13 UTC
2e8cbb58f395ea4546399d3d019e38cf4d09c3cd
1 file changed
+1
-1
vllm/v1/worker/gpu_model_runner.py
+1
-1
@@ -2079,7 +2079,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
2079
block_table_tensor=self.input_batch.block_table[
2080
kv_cache_group_id].get_device_tensor()[:num_reqs],
2081
slot_mapping=self.input_batch.
2082
- block_table[kv_cache_group_id].slot_mapping[:num_reqs])
2082
+ block_table[kv_cache_group_id].slot_mapping[:num_tokens])
2083
2084
attn_metadata_i = self.attn_metadata_builders[
2085
kv_cache_group_id].build_for_cudagraph_capture(