[CI] Add more ported distributed cases #2082
Conversation
zxd1997066
force-pushed
the
xiangdong/ported_cases
branch
13 times, most recently
from
0d9b54f to
85fa6f1
Compare
|
|
||
| inputs: | ||
| ut_name: | ||
| required: true |
There was a problem hiding this comment.
| required: true | |
| required: false |
| ze = xpu_list[i+1]; | ||
| } else { | ||
| ze = i; | ||
| if [ "${{ inputs.ut_name }}" == "xpu_distributed" ];then |
There was a problem hiding this comment.
Is there any assumptions in here? Can we detect topology directly and dynamically on the test node?
Please consider below scenarios:
- No Xelink group, return failed
- 1 Xelink group, launch 1 worker
- 2 Xelink group, launch 2 workers
- ...
.github/workflows/_linux_ut.yml
Outdated
| runner: | ||
| runs-on: ${{ inputs.runner }} | ||
| name: get-runner | ||
| name: get-runner |
There was a problem hiding this comment.
why we have such change?
zxd1997066
force-pushed
the
xiangdong/ported_cases
branch
14 times, most recently
from
3ac75d3 to
82b8ed3
Compare
zxd1997066
force-pushed
the
xiangdong/ported_cases
branch
5 times, most recently
from
63799f6 to
1bcbce2
Compare
firstly added cases for CI in this PR, will enable nightly test in another PR |
zxd1997066
force-pushed
the
xiangdong/ported_cases
branch
from
1bcbce2 to
87e933c
Compare
|
Seems some nodes have issue when running distributed test on the second group worker, like dut7362 https://github.com/intel/torch-xpu-ops/actions/runs/18654925169/job/53261876278#step:4:741, there are extra failures, and stop due to unknown issue. A previous good node is gm03pappvc004 https://github.com/intel/torch-xpu-ops/actions/runs/18632171421/job/53144659640 |
2 participants
This PR intends to add more ported distributed cases in torch-xpu-ops CI. And add pytest-xdist for distributed UT
The distributed UT time will increase to 1h22min with 2 work groups https://github.com/intel/torch-xpu-ops/actions/runs/18632171421/job/53144659640
disable_e2e
disable_ut