Lukasz Stafiniak
Lukasz Stafiniak
Or maybe not, for the simplest "k+1 dimensional" cases there's a workaround using fixed indexing.
Bumping this, combined with retrieving settings via `get_global_arg` (which checks commandline, environment vars, and a config file).
This shouldn't be the _default_ settings, but the settings retrieved by `get_global_arg`.
We have great support for settings now, with settings files.
This is not a viable backend candidate (not flexible enough).
Have a story about / support for experiment tracking: graphs of observables e.g. loss, device health
I thought of supporting numerical logging in ppx_minidebug, it's one direction of resolving this issue.
Have a story about / support for experiment tracking: graphs of observables e.g. loss, device health
W&B is the gold standard. Some ideas outside the Weights and Biases space: - [Ketrew: Keep Track of Experimental Workflows](https://github.com/hammerlab/ketrew), not machine-learning specific, but an OCaml library already, might turn...
https://github.com/johnma2006/mamba-minimal
The refactor is done for a long time now, but there's plenty work to make `arrayjit` a viable library -- dedicated tests, printing, possibly refactoring shape inference also? Not sure...
It's not an independent package, but it is a separate opam package... That's good enough. The original issue title does not say "independent".