[Performance] Make safetensors checkpoint prefetch settings configurable (#41499)

Signed-off-by: Mohammad Miadh Angkad <MAngkad.BSDSBA2027@aim.edu>
This commit is contained in:
Mohammad Miadh Angkad
2026-05-10 15:55:15 +00:00
committed by GitHub
parent f396bee56f
commit 21943d4c25
4 changed files with 87 additions and 15 deletions
@@ -256,6 +256,12 @@ class DefaultModelLoader(BaseModelLoader):
self.load_config.use_tqdm_on_load,
self.load_config.safetensors_load_strategy,
local_expert_ids=self.local_expert_ids,
safetensors_prefetch_num_threads=(
self.load_config.safetensors_prefetch_num_threads
),
safetensors_prefetch_block_size=(
self.load_config.safetensors_prefetch_block_size
),
)
else:
if extra_config.get("enable_multithread_load"):