sci-ml/pytorch-metric-learning
Loss functions, samplers, and trainers for metric learning in PyTorch
ChangeLog
commit 6f193ffc328459a4863423d293cebfd763ec1476
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Nov 1 12:26:00 2026 +0100
sci-ml/pytorch-metric-learning: add 3.0.1
Version 3 removes the trainers and testers, and with them the hard
dependencies on scikit-learn, scipy and tqdm; only numpy and pytorch
remain, at higher floors. scikit-learn and faiss are loaded on demand.
2.9.0 stays for pyannote-audio, which is capped below 3.
commit d4a034e227685625a04153b0b3f75fa941db1b17
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 07:01:30 2026 +0200
sci-ml/pytorch-metric-learning: trim 2.9.0 dependency rationale
commit c1916ad8c2ce50edebc479ca5ca5954646884a87
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Jun 16 22:30:36 2026 +0200
sci-ml/pytorch-metric-learning: Keyword 2.9.0 for ~arm64
Pure-Python (PEP517), architecture-independent.
commit 574dabbae81e5ddba02283de63003bf126eef7ee
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat May 16 10:09:09 2026 +0200
sci-ml/pytorch-metric-learning: new package, add 2.9.0
Loss functions, samplers, and trainers for metric learning in PyTorch.
Required runtime dep of sci-ml/pyannote-audio (speaker-embedding
training paths).
Pre-PEP-625 sdist preserves dashes in filename; PYPI_NO_NORMALIZE
keeps the URL working. SINGLE_IMPL via pytorch. Tests restricted
(faiss + tensorboard fixtures).
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Nov 1 12:26:00 2026 +0100
sci-ml/pytorch-metric-learning: add 3.0.1
Version 3 removes the trainers and testers, and with them the hard
dependencies on scikit-learn, scipy and tqdm; only numpy and pytorch
remain, at higher floors. scikit-learn and faiss are loaded on demand.
2.9.0 stays for pyannote-audio, which is capped below 3.
commit d4a034e227685625a04153b0b3f75fa941db1b17
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 07:01:30 2026 +0200
sci-ml/pytorch-metric-learning: trim 2.9.0 dependency rationale
commit c1916ad8c2ce50edebc479ca5ca5954646884a87
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Jun 16 22:30:36 2026 +0200
sci-ml/pytorch-metric-learning: Keyword 2.9.0 for ~arm64
Pure-Python (PEP517), architecture-independent.
commit 574dabbae81e5ddba02283de63003bf126eef7ee
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat May 16 10:09:09 2026 +0200
sci-ml/pytorch-metric-learning: new package, add 2.9.0
Loss functions, samplers, and trainers for metric learning in PyTorch.
Required runtime dep of sci-ml/pyannote-audio (speaker-embedding
training paths).
Pre-PEP-625 sdist preserves dashes in filename; PYPI_NO_NORMALIZE
keeps the URL working. SINGLE_IMPL via pytorch. Tests restricted
(faiss + tensorboard fixtures).


View
Download
Browse