I’m going to use meta_schedule to tune my operator. But I don’t find any API docs available. Do we have MetaSchedule Python API Docs? I searched at https://tvm.apache.org/docs/v0.11.0/reference/api/python, but find nothing about MetaSchedule. Please help
Maybe Unittest of MetaSchedule helps. That’s the way I got familiar with MS now.
OK, thank you. I will try it
@Hzfengsy i think @MasterJianxing is asking for docs rather than tutorials. it’s weird that MetaSchedule docs are not showing up there though
Either tutorials or MetaSchedule Docs is OK. But now we still have nothing to read and learn
Looks like meta_schedule docs were added with PR #14480, and I can see the docs available now at tvm.meta_schedule — tvm 0.12.dev0 documentation
thanks @zxybazh for your effort getting MetaSchedule docs online!
I looked into the docs, however, it seems all secondary APIs are not revealed for some reason, for example, tvm.meta_schedule.runner.LocalRunner
. Not sure what to do
Hello @junrushao ,
As you said, secondary APIs have not been revealed yet in the metascheduling docs.
Also, will it be possible to get a tutorial on how to use metascheduling? I understand that the relevant code has already been upstreamed to the main tvm and I am able to locate the metascheduling
folder in tvm, but the learning it via unittests alone is a bit tedious.
Similar to how AutoTVM and AutoScheduler have tutorials, for example - Auto-tuning a Convolutional Network for NVIDIA GPU — tvm 0.15.dev0 documentation (apache.org), please include an end to end tutorial for meta scheduling as well. Thank you so much. @zxybazh
Yes, a tutorial would be a huge lifesaver at this point.
Hi @Krishna, thanks for the suggestion! Documentations are under developement and it’s definitely something we want to have. For now I think the MLC tutorial would be a good start to get familiar with MS tuning. And I also shared some tips in the other thread that you were asking.
Also please feel free to create another thread for specific question or contribute to a tutorial if you’d like. Thanks a lot!
Hi @zxybazh ,
Thank you so much for your quick response. I will definitely check these resources first. I will make sure to open a new thread for any further queries and suggestions reg this issue.
Thanks again!