[Fast Start] Preload the FlashInfer autotune table on the weight cache daemon (#60085)

Signed-off-by: liusy58 <mg21330037@smail.nju.edu.cn>
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
This commit is contained in:
siyu
2026-10-10 15:48:06 +00:00
committed by GitHub
co-authored by Claude Opus 5 Isotr0py
parent 3709632ff2
commit f2376cdc4d
10 changed files with 253 additions and 17 deletions
+9 -3
View File
@@ -373,9 +373,15 @@ class GPUModelRunner(LoRAModelRunnerMixin):
tasks.extend(PoolingRunner.get_supported_tasks(self.model))
return tuple(tasks)
def load_model(self, load_dummy_weights: bool = False, *args, **kwargs) -> None:
def load_model(
self,
load_dummy_weights: bool = False,
*args,
model: nn.Module | None = None,
**kwargs,
) -> None:
time_before_load = time.perf_counter()
if load_dummy_weights:
if model is None and load_dummy_weights:
self.load_config.load_format = "dummy"
self.eplb.prepare_load()
eplb_models_added = False
@@ -385,7 +391,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
# Capture warmup providers selected while constructing the model.
with self.jit_warmup_registry.activate():
self.model = model_loader.load_model(
self.model = model or model_loader.load_model(
vllm_config=self.vllm_config,
model_config=self.vllm_config.model_config,
)