search

pytorch / torchx

TorchX is a universal job launcher for PyTorch applications. TorchX is designed to have fast iteration time for training/research and support for E2E production ML pipelines when you're ready.
https://pytorch.org/torchx
Other
326 stars 109 forks source link

[torchx/config] Generate docs on the available configuration options in .torchxconfig #375

Open kiukchung opened 2 years ago

kiukchung commented 2 years ago

📚 Documentation

Note: not a request for correction of documentation!

Link

https://pytorch.org/torchx/latest/experimental/runner.config.html

What does it currently say?

Nothing wrong with the current docs, but would be nice to have a list of the options that are "set-able" via .torchxconfig

What should it say?

Add a section that lists out the possible options and section names. Note that some options (e.g. the types of schedulers available and their respective runopts) are different between Meta-internal and OSS. Having a contextual TorchXConfig DAO-like object with placeholders and generating a docs page by dumping that object would make it possible to capture these differences.

Why?

Currently it is not clear what options can/cannot be set via .torchxconfig, we need a glossary of all the available options along with a help string on what they do and default values (if any)

kiukchung commented 2 years ago

FWIW this PR adds scheduler runoptions to each scheduler docs page by using SphinxDirective to query the runopts from the schedulers and generating the docs page. https://github.com/pytorch/torchx/pull/374

d4l3k commented 2 years ago

The documentation here has improved a fair amount, not sure if this is still a relevant issue https://pytorch.org/torchx/main/runner.config.html