Conversation
|
|
||
| # Install packages for processing the performance results | ||
| RUN pip3 install --upgrade pip | ||
| RUN pip3 install sqlalchemy==1.4.46 |
There was a problem hiding this comment.
Put pip instlal in a RUN command and install sympy recordtype parameterized einops jinja2 too.
Also add pined lint python package pip install ufmt==2.0.1 click==8.1.3 black==22.12.0 flake8==5.0.4.
There was a problem hiding this comment.
Please go ahead and add any packages you need.
examples/run_tests.sh
Outdated
| print_log_header 04_vit.log $hostname $GIT_BRANCH | ||
| HIP_VISIBLE_DEVICES=0 python3 benchmark_ait.py 2>&1 | tee -a 04_vit.log | ||
| # test 2 gcd | ||
| for BATCH_SIZE in 1 2 4 8 16 32 64 128 256 |
There was a problem hiding this comment.
Maybe there should be less batch sizes? Such as 1, 8, 32, 256
There was a problem hiding this comment.
I'll discuss the number of tests with Jing and Chao tomorrow. Right now, when we run our CI on jenkins nodes, we cannot count on more than 1 GPU being there, so everything got pretty slow, 5-6 hours. I would recommend cutting down the number of tests to 2-3 cases for each test type on a standard CI and maybe run one full set on a daily basis with develop branch.
|
Why the PR pull to amd-develop branch instead of merge_upstream |
|
Because this is CI for our repo, so we need it in our repo, not in FB one. |
|
Thanks, I understand. |
|
Next time when we merge up from our repo to FB, we will omit the Jenkinsfile and just push the updated docker file and test scripts. |
2 participants
This commit includes several changes: