Files
spark-vllm-docker/fastsafetensors.patch

13 lines
633 B
Diff

diff --git a/vllm/model_executor/model_loader/weight_utils.py b/vllm/model_executor/model_loader/weight_utils.py
index d43656c4f382..7025efd1c2de 100644
--- a/vllm/model_executor/model_loader/weight_utils.py
+++ b/vllm/model_executor/model_loader/weight_utils.py
@@ -826,6 +826,7 @@ def fastsafetensors_weights_iterator(
pg = SingleGroup()
device = torch.device(f"cuda:{current_platform.current_device()}")
+ hf_weights_files = sorted(hf_weights_files, key=_natural_sort_key)
weight_files_sub_lists = [
hf_weights_files[i : i + pg.size()]
for i in range(0, len(hf_weights_files), pg.size())