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, 20:00 UTC
update torchao safetensors impl (#30155)
Signed-off-by: Angel Li <liangel@meta.com>
liangel-02 committed
Dec 8, 2025 at 23:46 UTC
4b03b502119f857bbead7064a29267b9ea8999e5
1 file changed
+1
-1
vllm/model_executor/model_loader/weight_utils.py
+1
-1
@@ -641,7 +641,6 @@ def safetensors_weights_iterator(
641
if safetensors_load_strategy == "eager":
642
loading_desc += " (eager)"
643
644
- state_dict = {}
644
leftover_state_dict: dict[str, torch.Tensor] = {}
645
646
for st_file in tqdm(
@@ -667,6 +666,7 @@ def safetensors_weights_iterator(
666
)
667
668
with safe_open(st_file, framework="pt") as f:
669
+ state_dict = {}
670
for name in f.keys(): # noqa: SIM118
671
state_dict[name] = f.get_tensor(name)
672