DzAvril commented on pull request #10650: URL: https://github.com/apache/tvm/pull/10650#issuecomment-1074701615
@masahi OK, I will send a Pre-RFC first in the discussion forum. @comaniac I think we have your use cases too. Our solution is mixing AutoTVM and auto-schedule to tune one model by adding a config `include_op_impls` which lists the certain anchor op with implementation named `conv2d_nchw_int8.cuda` for example. Then AutoTVM only takes charge of certain subgraphs with fixed anchor op listed in `include_op_impls` and auto-schedule takes charge of the others. Combine with the feature proposed in this PR, we can achieve much better performance than tuning with AutoTVM or auto-schedule individually. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
