[Fast Start] Charge daemon-held weights against gpu_memory_utilization (#57298)

Signed-off-by: liusy58 <mg21330037@smail.nju.edu.cn>
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Signed-off-by: Xun Sun <UNIDY2002@outlook.com>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: Xun Sun <UNIDY2002@outlook.com>
This commit is contained in:
siyu
2026-09-29 17:58:35 +00:00
committed by GitHub
co-authored by Isotr0py Xun Sun
parent cb4f016c5d
commit f5ed3d9c03
7 changed files with 187 additions and 14 deletions
+8 -1
View File
@@ -57,6 +57,7 @@ from vllm.distributed.weight_transfer import (
)
from vllm.logger import init_logger
from vllm.lora.request import LoRARequest
from vllm.model_executor.model_loader import get_model_loader
from vllm.model_executor.warmup.kernel_warmup import kernel_warmup
from vllm.multimodal.gpu_ipc_memory import reserve_mm_ipc_gpu_memory
from vllm.platforms import current_platform
@@ -487,7 +488,13 @@ class Worker(WorkerBase):
# take current memory snapshot
self.init_snapshot = init_snapshot = MemorySnapshot(device=self.device)
self.requested_memory = request_memory(init_snapshot, self.cache_config)
# Weights from external model loader process
external_weight_memory = get_model_loader(
self.load_config
).get_external_weight_memory(self.vllm_config)
self.requested_memory = request_memory(
init_snapshot, self.cache_config, external_weight_memory
)
logger.debug("worker init memory snapshot: %r", self.init_snapshot)
logger.debug(
"worker requested memory: %sGiB", format_gib(self.requested_memory)