dev-python/flashinfer-cubin
Pre-compiled cubins for FlashInfer kernels
Screenshots
View
Download
BrowseChangeLog
commit eb22ce35bf24a0caa261fe42967b3cb435739d7d
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 7 15:33:12 2026 +0200
dev-python/flashinfer-cubin: new package, 0.6.8.post1
Tier 0 leaf for the vllm CUDA target packaging cycle. Pre-compiled
NVIDIA cubins for FlashInfer kernels — shipped only as a single
py3-none-any wheel on PyPI; no upstream source. Required as a runtime
sidecar by flashinfer-python.
Gentoo's PMS version syntax forbids ".postN", so PyPI's "0.6.8.post1"
is encoded as Gentoo PV "0.6.8_p1" with MY_PV translating back to
the PyPI form for the wheel filename.
The wheel is large (~280 MiB) — pre-compiled cubins for many CUDA
arches. Marked QA_PREBUILT and bindist-restricted.
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 7 15:33:12 2026 +0200
dev-python/flashinfer-cubin: new package, 0.6.8.post1
Tier 0 leaf for the vllm CUDA target packaging cycle. Pre-compiled
NVIDIA cubins for FlashInfer kernels — shipped only as a single
py3-none-any wheel on PyPI; no upstream source. Required as a runtime
sidecar by flashinfer-python.
Gentoo's PMS version syntax forbids ".postN", so PyPI's "0.6.8.post1"
is encoded as Gentoo PV "0.6.8_p1" with MY_PV translating back to
the PyPI form for the wheel filename.
The wheel is large (~280 MiB) — pre-compiled cubins for many CUDA
arches. Marked QA_PREBUILT and bindist-restricted.