Fast Start (#54921)

Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: liusy58 <liusy58@smail.nju.edu.cn>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
This commit is contained in:
siyu
2026-09-04 06:32:06 -07:00
committed by GitHub
co-authored by liusy58 Isotr0py
parent 07ee97e023
commit a69e75b9b6
16 changed files with 1529 additions and 13 deletions
@@ -27,6 +27,13 @@ class QuantizeMethodBase(ABC):
in process_weights_after_loading, reducing peak memory during loading.
"""
supports_pre_processed_weights: bool = False
"""
Whether ``process_weights_after_loading`` supports running under
``weights_already_processed``. Methods must skip tensor transforms in
that mode; the loader driver rejects methods that do not declare support.
"""
@abstractmethod
def create_weights(
self, layer: torch.nn.Module, *weight_args, **extra_weight_attrs