DickJC123 opened a new pull request #19296:
URL: https://github.com/apache/incubator-mxnet/pull/19296
## Description ##
This PR includes improvements discovered during the testing of the v1.8.x
branch:
- For the test_tensorrt.py:test_tensorrt_symbol test, the explicitly set
tolerances were removed for the 32-bit model test. This allows the test to
pass on A100 by allowing the framework to detect the A100 context and choose
the relaxed tolerance appropriate for a TF32 calculation.
- For the test_tensorrt.py:test_tensorrt_symbol_int8 test, logic was
inserted to avoid running the test on GPU architectures earlier than Volta
(PASCAL, MAXWELL, etc.). On PASCAL, where int8 is not supported, a message was
printed, but then the test hung.
With this PR, the test is skipped. In general, users should not attempt
running TRT int8 on PASCAL GPUs or earlier.
- Finally, a CI failure was seen in
test_operator_gpu.py:test_np_mixed_precision_binary_funcs for seed
MXNET_TEST_SEED=590323404. Additional repro'd failures all seemed to occur
when the data shape was (). This behavior was seen and fixed in master PR
https://github.com/apache/incubator-mxnet/pull/18660. This PR includes only
the portion of that PR that skips the test when the data shape is (). The full
PR should still be backported by the author to 1.x if deemed useful.
@yzhliu @Kh4L @samskalicky
## Checklist ##
### Essentials ###
- [X ] PR's title starts with a category (e.g. [BUGFIX], [MODEL],
[TUTORIAL], [FEATURE], [DOC], etc)
- [X ] Changes are complete (i.e. I finished coding on this PR)
- [X ] All changes have test coverage
- [X ] Code is well-documented
### Changes ###
- [ ] Feature1, tests, (and when applicable, API doc)
- [ ] Feature2, tests, (and when applicable, API doc)
## Comments ##
- If this change is a backward incompatible change, why must this change be
made.
- Interesting edge cases to note here
----------------------------------------------------------------
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:
[email protected]
