Install this version:
emerge -a =dev-python/tokenspeed-mla-bin-0.2.5If this version is masked, you can unmask it using the autounmask tool or standard emerge options:
autounmask =dev-python/tokenspeed-mla-bin-0.2.5Or alternatively:
emerge --autounmask-write -a =dev-python/tokenspeed-mla-bin-0.2.5| Version | EAPI | Keywords | Slot |
|---|---|---|---|
| 0.2.5 | 8 | ~amd64 | 0 |
# Copyright 1999-2026 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2
EAPI=8
DISTUTILS_USE_PEP517=no
DISTUTILS_SINGLE_IMPL=1
PYTHON_COMPAT=( python3_{12..14} )
inherit distutils-r1
MY_WHEEL="tokenspeed_mla-${PV}-py3-none-manylinux_2_28_x86_64.whl"
DESCRIPTION="TokenSpeed multi-head latent attention CUDA kernels"
HOMEPAGE="https://pypi.org/project/tokenspeed-mla/"
SRC_URI="https://files.pythonhosted.org/packages/68/50/de595f2df20d52d2e12df5c2032a715a5ee07ae32f120b96e231b67dcbee/${MY_WHEEL}"
S=${WORKDIR}
LICENSE="MIT"
SLOT="0"
KEYWORDS="~amd64"
RESTRICT="strip"
RDEPEND="
$(python_gen_cond_dep '
>=dev-python/apache-tvm-ffi-0.1.13[${PYTHON_USEDEP}]
<=dev-python/apache-tvm-ffi-0.1.13-r0[${PYTHON_USEDEP}]
dev-python/nvidia-cutlass-dsl[${PYTHON_USEDEP}]
')
>=dev-python/tokenspeed-triton-bin-3.8.10_p20260721[${PYTHON_SINGLE_USEDEP}]
sci-ml/caffe2
"
QA_PREBUILT="usr/lib*/python*/site-packages/tokenspeed_mla/objs/*.so*"
python_install() {
${EPYTHON} -m installer --destdir="${D}" "${DISTDIR}/${MY_WHEEL}" || die
}
$(python_gen_cond_dep ' >=dev-python/apache-tvm-ffi-0.1.13[${PYTHON_USEDEP}] <=dev-python/apache-tvm-ffi-0.1.13-r0[${PYTHON_USEDEP}] dev-python/nvidia-cutlass-dsl[${PYTHON_USEDEP}] ') >=dev-python/tokenspeed-triton-bin-3.8.10_p20260721[${PYTHON_SINGLE_USEDEP}] sci-ml/caffe2