Refactor monarch variable names to boost readbility

[Edenzzzz]

Previously block dimensions such as q, p are very non-intuitive and make the code hard to understand and build on. In developing my Monarch PEFT project, I renamed them to fix the block-diagonal matmul taxonomy with terms like blk_in, blk_out to clearly indicate block dims.
This could significantly encourage follow-up work using this code base, as folks have complained about hard-to-use code :)

[Edenzzzz]

Precision tests passed in my repo @DanFu09 https://github.com/SprocketLab/sparse_matrix_fine_tuning/blob/main/tests/ops/test_blockdiag_butterfly_multiply.py

[DanFu09]

Cool! Has this been tested in this repo in any models?

[Edenzzzz]

If you mean convergence tests, not yet--only passed precision tests. Can probably try that in a few days. Thanks!

[DanFu09]

Thank you! Just some basic curves should be fine — in case there’s some end-to-end disruption.

…

On Sun, Dec 15, 2024 at 9:49 AM Wenxuan Tan ***@***.***> wrote: If you mean convergence tests, not yet--only passed precision tests. Can probably try that in a few days. Thanks! — Reply to this email directly, view it on GitHub <#40 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABDDIIWBWZV34ZL2JMF4GM32FW6LTAVCNFSM6AAAAABTRAQ73CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNBTHE3TIMZRGE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[Edenzzzz]

@DanFu09 Convergence curves match (monarch-mixer-finetune-glue-768dim-80m-parameters.yaml)