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 8, 15:15 UTC
[Bugfix] Sort hf_weights_files in fastsafetensors_weights_iterator to match #33491 (#34190)
Signed-off-by: Balaxxe <136368465+jaim12005@users.noreply.github.com>
Balaxxe committed
Feb 10, 2026 at 00:06 UTC
8d48d0a9d9edfc2eb9cee6bb941be20211eb8282
1 file changed
+1
vllm/model_executor/model_loader/weight_utils.py
+1
@@ -826,6 +826,7 @@ def fastsafetensors_weights_iterator(
826
pg = SingleGroup()
827
828
device = torch.device(f"cuda:{current_platform.current_device()}")
829
+ hf_weights_files = sorted(hf_weights_files, key=_natural_sort_key)
830
weight_files_sub_lists = [
831
hf_weights_files[i : i + pg.size()]
832
for i in range(0, len(hf_weights_files), pg.size())