Skip to content

Conversation

hameerabbasi
Copy link

No description provided.

@hameerabbasi hameerabbasi requested a review from amjames June 27, 2025 17:31
@amjames
Copy link
Member

amjames commented Jul 17, 2025

@hameerabbasi I am holding this for now pending a discussion about the future of our tooling in here.

My thoughts:

  • I don't really want to collect a bunch of stuff here to support the ins and outs of how everyone prefers to work. This should be a simple one stop shop to get anyone set up for development quickly, and it should support the variations on builds/environment that we need to cover the development we do.
  • I would support going all in on Pixi, and making it the main driver for this system, refactor/remove some of our scripts and move forward with that.

@hameerabbasi
Copy link
Author

This ideally requires pytorch/benchmark#2622 to be merged first.

One caveat here is that torchbenchmark mutates the environment; its dependencies aren't consistent; and so the environment may end up becoming inconsistent with the lock file.

One solution is to have a separate benchmark environment that can be mutated without affecting the others.

@hameerabbasi
Copy link
Author

This should be a simple one stop shop to get anyone set up for development quickly, and it should support the variations on builds/environment that we need to cover the development we do.

Done! It now supports all the YAML files in the original repo.

I would support going all in on Pixi, and making it the main driver for this system, refactor/remove some of our scripts and move forward with that.

I would too, the only caveat being in #34 (comment).

Comment on lines +10 to +11
echo "source $SCRIPT_DIR/torch-common.sh" > $CONDA_PREFIX/etc/conda/activate.d/activate-torch.sh
echo "source $SCRIPT_DIR/deactivate-torch-common.sh" > $CONDA_PREFIX/etc/conda/activate.d/deactivate-torch.sh
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is needed to make sure the environment is set up properly

Copy link
Author

@hameerabbasi hameerabbasi Jul 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

With conda, yes. But if you look at the following lines; pixi will take care of that for us. I'm calling the script inside the activation.scripts section of pixi.toml.

torch-build/pixi.toml

Lines 18 to 19 in 9634e20

[activation]
scripts = ["./torch-common.sh"]

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What about deactivation?

Copy link
Author

@hameerabbasi hameerabbasi Jul 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pixi tasks are generally run under a new shell process with pixi shell or pixi run instead of in the same process. Thus the deactivation cleanup isn't needed. One CAN use the same shell with eval "$(pixi shell-hook)", but that's a fringe use-case.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants