Loading Model weights with fastsafetensorsΒΆ
Using fastsafetensors library enables loading model weights to GPU memory by leveraging GPU direct storage. See their GitHub repository for more details.
To enable this feature, use the --load-format fastsafetensors
command-line argument