Forschungsteam vor Bildschirmen mit Visualisierungen künstlicher neuronaler Netze

Projekt

mushin: boilerplate-free, reproducible machine-learning experiment sweeps

mushin runs reproducible machine-learning experiment sweeps with minimal boilerplate: decorate a function with @mushin.sweep, declare the swept parameters, and get the results back as a labeled xarray.Dataset. Built on PyTorch Lightning and hydra-zen, it captures per-run provenance, resumes durably after interruption…

mushin runs reproducible machine-learning experiment sweeps with minimal boilerplate: decorate a function with @mushin.sweep, declare the swept parameters, and get the results back as a labeled xarray.Dataset. Built on PyTorch Lightning and hydra-zen, it captures per-run provenance, resumes durably after interruption (including cluster preemption), and scales from a laptop to a multi-node SLURM cluster by changing only the launcher.

Themengebiete