ci: print annotations for key package versions in transformers test #1184
Conversation
dvrogozh
force-pushed
the
ci
branch
7 times, most recently
from
366461a to
2a843d5
Compare
Signed-off-by: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
Annotations are available on a summary page of executed workflow. Signed-off-by: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
Signed-off-by: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
| echo "| jobs.$GITHUB_JOB.versions.python | $(python --version | cut -f2 -d' ') |" >> $GITHUB_STEP_SUMMARY | ||
| packages=" \ | ||
| level-zero \ | ||
| libigc1 \ |
There was a problem hiding this comment.
Note that this accounts to both cases: LTS and rolling drivers. They have some components available in the packages with the different names, like libigc1 vs. libigc2 and same for level-zero. This will result in printing empty versions for not installed packages.
|
LGTM. @chuanqi129 Could you please double check? |
| echo "pip installed packages:" | ||
| pip list | tee ${{ github.workspace }}/transformers/tests_log/pip_list.txt | ||
| echo "lspci gpu devices:" | ||
| lspci -d ::0380 | tee ${{ github.workspace }}/transformers/tests_log/lspci_0380.txt |
There was a problem hiding this comment.
Can we use xpu-smi discovery to get the gpu devices information? Because the runner procession with ZE_AFFINITY_MASK by default, the lspci always show all devices information.
There was a problem hiding this comment.
Good idea. But let's add running xpu-smi in a stand-alone step. I remember it gives lengthy output by itself and for us it has special interest. Good to separate from this step. I will submit another PR on this since I need to experiment with xpu-smi a little for the better output.
| echo "| jobs.$GITHUB_JOB.torch.xpu.device_count | $var |" >> $GITHUB_STEP_SUMMARY | ||
| # printing annotations with key environment variables | ||
| echo "| jobs.$GITHUB_JOB.env.ZE_AFFINITY_MASK | $ZE_AFFINITY_MASK |" >> $GITHUB_STEP_SUMMARY | ||
| echo "| jobs.$GITHUB_JOB.env.NEOReadDebugKeys | $NEOReadDebugKeys |" >> $GITHUB_STEP_SUMMARY |
There was a problem hiding this comment.
how do you think if we convert this step as a standalone scripts instead of long step in the workflow?
There was a problem hiding this comment.
It's not that long at the moment to strongly require converting to the script per my opinion.... Eventually complicated stuff, or often reused stuff, or lengthy stuff needs conversion to the script. At the same time I personally am trying to avoid calling scripts from github actions since this impacts workflow transparency and ultimately complicates workflow. So, right balance needs to be looked for here.
For this particular step, I actually wonder maybe we need to convert it to the stand-alone action rather than script? In this way it can be reused by other workflows in the repo in their steps and bring unified format for reporting this information.
Overall, let's merge in this PR as is and discuss these proposals.
|
PR did not modify the scope of failing ci test (pull/preci-ut/Torch-XPU-UT-Tests). Can be merged. |
Annotations are available on a summary page of executed workflow.