gpo.zugaina.org

Search Portage & Overlays:

dev-python/fastsafetensors

High-performance safetensors model loader (GPUDirect Storage)

Screenshots

  • fastsafetensors-0.3.2
    ~amd64
    python_targets_python3_12 python_targets_python3_13 python_targets_python3_14 debug

    View      Download      Browse     License: Apache-2.0   
    Overlay: stuff
  • fastsafetensors-0.3.1
    ~amd64
    python_targets_python3_12 python_targets_python3_13 python_targets_python3_14 debug

    View      Download      Browse     License: Apache-2.0   
    Overlay: stuff
  • fastsafetensors-0.2.2
    ~amd64
    python_targets_python3_12 python_targets_python3_13 python_targets_python3_14 debug

    View      Download      Browse     License: Apache-2.0   
    Overlay: stuff

ChangeLog

commit 83f3cef2be9736e9252baf052ebace6378ba4677
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat May 23 00:39:56 2026 +0200

dev-python/fastsafetensors: add 0.3.2

commit 0e0067b44d5604d3784f5c005650e34ccfeac497
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 13 14:34:45 2026 +0200

dev-python/fastsafetensors: disable py3.11

commit ab80e1109b735162df39f98f8b821875600898d6
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 10 15:14:32 2026 +0200

dev-python/fastsafetensors: add 0.3.1

commit 071260b9634f526939edd76b2da6388dfa588b90
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 7 14:48:43 2026 +0200

dev-python/fastsafetensors: new package, 0.2.2

Tier 0 leaf for the vllm CUDA target packaging cycle. High-performance
safetensors model loader using NVIDIA GPUDirect Storage (CUFile) for
direct disk-to-GPU paths.

The C++ extension is built with pybind11 and links only against stdc++
— libcuda / libcudart / libcufile are dlopen'd at runtime through the
self-contained cuda_compat.h shim, so no CUDA headers are needed at
build time and no link-time CUDA dep. nvidia-cuda-toolkit is only
required to actually exercise the GDS path; the CPU fallback works
without it, so it is omitted from RDEPEND — the consuming vllm CUDA
target ebuild already pulls the toolkit.