Pax is a Jax-based machine learning framework for training large scale models. Pax allows for advanced and fully configurable experimentation and parallelization, and has demonstrated industry leading model flop utilization rates.
Apache License 2.0
446
stars
68
forks
source link
[NVIDIA] Add synthetic configs for benchmarking #61
This PR adds synthetic dataset versions of the configs in configs.py for benchmarking, adds an accompanying bash script, and makes a few other minor fixes to some contrib/gpu scripts.
This PR adds synthetic dataset versions of the configs in
configs.pyfor benchmarking, adds an accompanying bash script, and makes a few other minor fixes to somecontrib/gpuscripts.