You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2020/11/16 18:15:10 UTC
[GitHub] [incubator-tvm] bernhardklein opened a new issue #6922: [AutoTVM][RPCRunner] timeout is not passed correctly
bernhardklein opened a new issue #6922:
URL: https://github.com/apache/incubator-tvm/issues/6922
In the [RPCRunner class](https://github.com/apache/incubator-tvm/blob/main/python/tvm/autotvm/measure/measure.py) the timeout is not correctly passed to the [LocalExecutor](https://github.com/apache/incubator-tvm/blob/main/python/tvm/autotvm/measure/local_executor.py) and therefore the default timeout is always used.
see previous problem description at [TVM discuss](https://discuss.tvm.apache.org/t/autotvm-dense-crashes-for-larger-matrix-sizes/8399)
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-tvm] comaniac commented on issue #6922: [AutoTVM][RPCRunner] timeout is not passed correctly
Posted by GitBox <gi...@apache.org>.
comaniac commented on issue #6922:
URL: https://github.com/apache/incubator-tvm/issues/6922#issuecomment-728240616
Looks like we need `self.executor = LocalExecutor(timeout=self.timeout)` in https://github.com/apache/incubator-tvm/blob/main/python/tvm/autotvm/measure/measure_methods.py#L243, but I'm curious why we never see `all_costs=120`, where 120 is the default timeout, in the tuning logs.
cc @merrymercy @FrozenGene @kevinthesun
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-tvm] bernhardklein commented on issue #6922: [AutoTVM][RPCRunner] timeout is not passed correctly
Posted by GitBox <gi...@apache.org>.
bernhardklein commented on issue #6922:
URL: https://github.com/apache/incubator-tvm/issues/6922#issuecomment-728243663
I created a matching pull request [6924](https://github.com/apache/incubator-tvm/pull/6924) for this.
We can not see it in the logs, I checked it in the logs is the correct highl-level value but the LocalExectuor works with the inner default. Probably there the read from the remote args works?
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-tvm] bernhardklein commented on issue #6922: [AutoTVM][RPCRunner] timeout is not passed correctly
Posted by GitBox <gi...@apache.org>.
bernhardklein commented on issue #6922:
URL: https://github.com/apache/incubator-tvm/issues/6922#issuecomment-728247695
Is not readable in the logs, because for the logs self.timeout is used, which is correct.
So with the pull request it should be ok.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-tvm] merrymercy closed issue #6922: [AutoTVM][RPCRunner] timeout is not passed correctly
Posted by GitBox <gi...@apache.org>.
merrymercy closed issue #6922:
URL: https://github.com/apache/incubator-tvm/issues/6922
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org