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, 03:30 UTC
[Misc] Fix `Unable to detect current VLLM config. Defaulting to NHD kv cache layout` warning (#20400)
Signed-off-by: NickLucche <nlucches@redhat.com>
Nicolò Lucchesi committed
Jul 3, 2025 at 23:56 UTC
8d775dd30a1461ffdb2ce019dff0bbed9c547ae8
2 files changed
+3
-3
vllm/distributed/kv_transfer/kv_connector/utils.py
+2
-2
@@ -97,10 +97,10 @@ def get_kv_connector_cache_layout():
97
# used for faster transfer.
98
vllm_config = get_current_vllm_config()
99
kv_config = vllm_config.kv_transfer_config
100
- if vllm_config.model_config is None or kv_config is None:
100
+ if kv_config is not None and vllm_config.model_config is None:
101
logger.warning_once("Unable to detect current VLLM config. " \
102
"Defaulting to NHD kv cache layout.")
103
- else:
103
+ elif kv_config is not None:
104
use_mla = vllm_config.model_config.use_mla
105
if not use_mla and kv_config.kv_connector == "NixlConnector":
106
logger.info_once("NixlConnector detected. Setting KV cache " \
vllm/v1/attention/backends/utils.py
+1
-1
@@ -138,7 +138,7 @@ def get_kv_cache_layout():
138
if cache_layout is None:
139
cache_layout = get_kv_connector_cache_layout()
140
else:
141
- logger.info_once("`FLASHINFER_KV_CACHE_LAYOUT` environment variable " \
141
+ logger.info_once("`VLLM_KV_CACHE_LAYOUT` environment variable " \
142
"detected. Setting KV cache layout to %s.", cache_layout)
143
144
return cache_layout