sync/ports
1
0
Fork 0
mirror of https://git.freebsd.org/ports.git synced 2025-04-28 09:36:41 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

misc/py-schedulefree: New port: Schedule free learning in PyTorch

Browse source
This commit is contained in:
Yuri Victorovich 2025-03-28 23:48:56 -07:00
parent 8866cf92ae
commit d9e91d6828
4 changed files with 40 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
misc/Makefile
View file

@ -513,6 +513,7 @@
SUBDIR += py-pytorch-lightning
SUBDIR += py-pyviz-comms
SUBDIR += py-safetensors
SUBDIR += py-schedulefree
SUBDIR += py-scikit-fusion
SUBDIR += py-scikit-quant
SUBDIR += py-serverfiles

25
misc/py-schedulefree/Makefile Normal file
View file

@ -0,0 +1,25 @@
PORTNAME= schedulefree
DISTVERSION= 1.4.1
CATEGORIES= misc python # machine-learning
MASTER_SITES= PYPI
PKGNAMEPREFIX= ${PYTHON_PKGNAMEPREFIX}
MAINTAINER= yuri@FreeBSD.org
COMMENT= Schedule free learning in PyTorch
WWW= https://github.com/facebookresearch/schedule_free
LICENSE= APACHE20
LICENSE_FILE= ${WRKSRC}/LICENSE
BUILD_DEPENDS= ${PYTHON_PKGNAMEPREFIX}hatchling>0:devel/py-hatchling@${PY_FLAVOR}
RUN_DEPENDS= ${PYTHON_PKGNAMEPREFIX}pytorch>0:misc/py-pytorch@${PY_FLAVOR} \
${PYTHON_PKGNAMEPREFIX}typing-extensions>0:devel/py-typing-extensions@${PY_FLAVOR}
USES= python
USE_PYTHON= pep517 autoplist pytest
NO_ARCH= yes
# most tests fail: AssertionError: Torch not compiled with CUDA enabled, see https://github.com/facebookresearch/schedule_free/issues/65
.include <bsd.port.mk>

3
misc/py-schedulefree/distinfo Normal file
View file

@ -0,0 +1,3 @@
TIMESTAMP = 1742853003
SHA256 (schedulefree-1.4.1.tar.gz) = 69ef25601d1fc0d8dd00cb36f9af78833f88b7846f1bb6ddecc9f144f3e9f7cb
SIZE (schedulefree-1.4.1.tar.gz) = 29281

11
misc/py-schedulefree/pkg-descr Normal file
View file

@ -0,0 +1,11 @@
Schedulefree is a Schedule-Free optimizer in PyTorch.
We provide several Schedule-Free optimizer implementations:
* SGDScheduleFree and SGDScheduleFreeReference: Schedule-free variants of SGD
* AdamWScheduleFree and AdamWScheduleFreeReference: Schedule-free variants
of AdamW
* RAdamScheduleFree: Schedule-free variant of RAdam, which eliminates the need
for both learning rate scheduling and warmup (implementation community
contributed)
* Experimental ScheduleFreeWrapper to combine with other optimizers