Conversation
haoyuz
force-pushed
the
haoyuzhang-xla-test
branch
from
a0e3d81 to
bdbdac4
Compare
haoyuz
changed the title
|
I will look at this tomorrow / Friday. I want to make sure the monkey patch does not mess up other tests or we run in isolation. I want to get this in ASAP. |
|
@tfboyd Sure! Maybe good to isolate XLA tests, or we can call |
tfboyd
left a comment
tfboyd
left a comment
There was a problem hiding this comment.
LGTM. These will run with all the other tests at first and I will segregate them if I have time next week.
| self._run_and_report_benchmark() | ||
|
|
||
| def benchmark_xla_8_gpu(self): | ||
| self._setup() |
There was a problem hiding this comment.
Need doc string to fix LINT issue.
| self._run_and_report_benchmark() | ||
|
|
||
| def benchmark_graph_xla_8_gpu(self): | ||
| self._setup() |
There was a problem hiding this comment.
Need doc string to fix LINT issue.
| FLAGS.batch_size = 128 | ||
| self._run_and_report_benchmark() | ||
|
|
||
| def benchmark_graph_xla_1_gpu(self): |
There was a problem hiding this comment.
Add doc string to avoid future LINT issue.
| self._run_and_report_benchmark() | ||
|
|
||
| def benchmark_xla_1_gpu(self): | ||
| self._setup() |
There was a problem hiding this comment.
Add doc string to avoid future LINT issue.
haoyuz
force-pushed
the
haoyuzhang-xla-test
branch
from
bdbdac4 to
53dcca7
Compare
|
@tfboyd Added doc strings to fix lint errors, PTAL. Thanks! |
tfboyd
left a comment
tfboyd
left a comment
There was a problem hiding this comment.
LGTM will wait for tests to run then merge.
3 participants
Add Keras XLA benchmarks and monkey-patched the
assert_broadcastableop to avoid OOM. The monkey patch should be reverted once the OOM issue is fixed.Tested with PerfZero on Google Cloud.