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 5, 14:45 UTC
[KVConnector] Skip `register_kv_caches` on profiling (#38558)
Signed-off-by: NickLucche <nlucches@redhat.com>
Nicolò Lucchesi committed
Apr 3, 2026 at 17:40 UTC
97f92c6b4741e99ac278f326cabef4f7884d0e49
1 file changed
+1
-1
vllm/v1/worker/gpu_model_runner.py
+1
-1
@@ -6844,7 +6844,7 @@ class GPUModelRunner(
6844
# group
6845
self.drafter.validate_same_kv_cache_group(kv_cache_config)
6846
6847
- if has_kv_transfer_group():
6847
+ if has_kv_transfer_group() and not is_profiling:
6848
kv_transfer_group = get_kv_transfer_group()
6849
if self.cross_layers_kv_cache is not None:
6850
assert self.cross_layers_attn_backend is not None