merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-717619470
How about also undoing the tutorials? I guess their format will be wrong
because you add indentation to text blocks.
For the autotvm problem, I am okay to accept the
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-717588164
I added `__name__ == "__main__"` but the results are the same.
Can you delete all autotvm related code and tutorials in this PR? We can
investigate later. In the
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-717550912
Sorry, could you point me to the related lines? I only see you rename some
variables in the `python/tvm/autotvm/measure/local_executor.py`
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-717547728
Which PR introduced the changes that broke the tutorials? I don't find
related thing in your PR.
This
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-717540915
What's your testing environment?
This is an automated message from the Apache Git Service.
To respond
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-717506384
I tested your autotvm implementation and found it is significantly slower.
To do the test, you can run this tutorial
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-716905146
The auto-scheduler part looks good to me.
But I am also not sure about the autotvm part.
If you don't mind, I suggest deleting the autotvm part in this PR.
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-716534539
Another concern is that I introduced a new kind of workload in this PR
(#6710) for relay integration.
When using auto-scheduler in Relay, we don't serialize a
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-716516782
Thanks for the refactoring! I would like to see benchmark results for both
autotvm and auto-scheduler, so we don't get performance regression.
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-708845921
Autotvm uses a lot of python multiprocessing and I expect it will be much
slower when using spawn. AutoTVM uses multiprocessing for feature extraction.
So it needs to
merrymercy commented on pull request #6671:
URL: https://github.com/apache/incubator-tvm/pull/6671#issuecomment-708103769
Note that "spawn" is slower than "fork"
(https://docs.python.org/3/library/multiprocessing.html#contexts-and-start-methods).
We should not change all default
11 matches
Mail list logo