I'm using a GitHub workflow to check my PyTorch-based monorepo with MyPy. One small issue: I'm running out of disk space. Am I misusing Pants? What are some common ways to get around this?

Based on the limited information provided that seems like a pretty normal way to use Pants. And you’re not using any GHA caching jobs, AFAICT, so this doesn’t seem like a cache GC issue.

Pytorch (from PyPi or the cuda specific wheels) pull in a lot of pre-compiled libraries, that add up to a few gigabytes -- likely duplicated by number of sandboxes. It could help setting

[python].run_against_entire_lockfile = true

in your pants.ci.toml -- I'm not sure if it would apply to mypy invocations. It would definitely in general improve caching.