Ray version upgrade #4302
Ray version upgrade #4302
Conversation
|
/benchmark module=tabular preset=tabular_medium benchmark=tabular_full time_limit=1h |
|
/benchmark module=tabular preset=tabular_medium benchmark=tabular_full time_limit=1h Benchmark Test Result - Pass Evaluation Results Path: s3://autogluon-ci-benchmark/evaluation/tabular/upgrade_ray_2.3x Benchmark Test Result - Fail Evaluation Results Path: s3://autogluon-ci-benchmark/evaluation/tabular/upgrade_ray_2.3x |
|
Job PR-4302-82339a9 is done. |
|
|
||
| def try_import_ray() -> ModuleType: | ||
| RAY_MAX_VERSION = "2.11.0" # sync with core/setup.py | ||
| RAY_MAX_VERSION = "2.40.0" # sync with core/setup.py |
There was a problem hiding this comment.
2.31 is the latest ray version, so this should be changed to "2.32"
core/setup.py
Outdated
| extras_require = { | ||
| "ray": [ | ||
| "ray[default]>=2.10.0,<2.11", # sync with common/src/autogluon/common/utils/try_import.py | ||
| "ray[default]>=2.10.0,<2.40", # sync with common/src/autogluon/common/utils/try_import.py |
There was a problem hiding this comment.
2.31 is the latest ray version, so this should be changed to "2.32"
core/setup.py
Outdated
| ], | ||
| "raytune": [ | ||
| "ray[default,tune]>=2.10.0,<2.11", # sync with common/src/autogluon/common/utils/try_import.py | ||
| "ray[default,tune]>=2.10.0,<2.40", # sync with common/src/autogluon/common/utils/try_import.py |
There was a problem hiding this comment.
2.31 is the latest ray version, so this should be changed to "2.32"
|
|
||
| def try_import_ray() -> ModuleType: | ||
| RAY_MAX_VERSION = "2.11.0" # sync with core/setup.py | ||
| RAY_MAX_VERSION = "2.40.0" # sync with core/setup.py |
There was a problem hiding this comment.
tabular_medium in AG-Bench I assume means medium_quality. medium_quality does not use Ray. We need to test best_quality which uses Ray.
There was a problem hiding this comment.
That's good to know. Let me run a benchmark with best_quality
There was a problem hiding this comment.
I have rerun the benchmark with best_quality, and the benchmark evaluation looks fine. There were some errors from ray after the time limit has reached, and I have rerun the benchmark locally to verify, but it turns out that the error seems to be sporadic, and happens on master as well.
This is the error message in the python-core*.log
Error reporting task backlog information: GrpcUnavailable: RPC Error message: failed to connect to all addresses; last error: UNKNOWN: ipv4:172.31.3.111:41673: Failed to connect to remote host: Connection refused; RPC Error details:
6628719 to
68ee339
Compare
39108cb to
1ad9ccc
Compare
|
Job PR-4302-1ad9ccc is done. |
Innixma
left a comment
There was a problem hiding this comment.
LGTM, thanks for running the benchmark!
Issue #, if available:
Description of changes:
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.