Is there an up-to-date uTVM code that I can refer to?

Great to hear! I think autotuning is unfortunately still busted at HEAD. My plan is to merge it behind the Project API refactor, since it introduced more Python dependencies to TVM I’d like to place in a separate dependency file.

Will there be a easy fix for me to enable autotuning, (e.g., is it going to be mainly patching the changed function call parameters) or is it something that might not be doable? If autotuning cannot be done with this repo, I am wondering if going back to your earlier experimental_blogpost branch would be better to try out autotuning, since you definitely had autotuning there (I already tried this actually – mbed-cli is not recognizing my device and I wonder if this is a rabbit hole worth digging).

Any suggestion on how I might be able to try out autotuning? Thank you always.

right now you’d need to merge PR 7545 and update python/micro_eval/bin/ to match. I did get autotuning working at a recent-ish (e.g. February) main witht PR 7545, so you could give it a shot if you’d like to try out autotuning now. Otherwise yeah, you may need to revert to the demo from last June :slightly_frowning_face: .

1 Like

I merged pin-tvm-repo into master, so there is now no need for the separate branch.

Thanks, I will first try merging this PR. Is it a PR for microTVM that simply hasn’t been merged? Also, are there any plan for making microTVM work again in master in the near future? Thanks!

hi @kwmaeng,

my post above was just following up to say that I have updated microtvm-blogpost-eval to work with approximately main as of Friday. I have pinned the working revision of apache/tvm repository in microtvm-blogpost-eval via git submodules, so, with the exception of autotuning, you shouldn’t see any breakage at that revision (please let me know if you do, so we can fix it).

Regarding PR 7545–yes, it enables autotuning with microTVM, and it will merge after Project API PR (branch) merges.