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 11, 21:15 UTC
[BugFix] Unblock use of LoRA with data parallel mode (#30220)
Signed-off-by: Nick Hill <nhill@redhat.com>
Nick Hill committed
Dec 7, 2025 at 20:21 UTC
d726a7b0ed0b58b7703709dc3e486c5fe2f55db3
1 file changed
+4
-1
vllm/v1/metrics/loggers.py
+4
-1
@@ -952,7 +952,10 @@ class PrometheusStatLogger(AggregateStatLoggerBase):
952
self.gauge_lora_info: Gauge | None = None
953
if vllm_config.lora_config is not None:
954
if len(self.engine_indexes) > 1:
955
- raise NotImplementedError("LoRA in DP mode is not supported yet.")
955
+ logger.warning(
956
+ "vllm:lora_requests_info prometheus metrics may be "
957
+ "incorrect/misleading with data parallel deployments."
958
+ )
959
self.labelname_max_lora = "max_lora"
960
self.labelname_waiting_lora_adapters = "waiting_lora_adapters"
961
self.labelname_running_lora_adapters = "running_lora_adapters"