Open TedThemistokleous opened 1 year ago
Just needs to pass review now
Still in review. Seeing failures due to Jenkins changes. Moving to current sprint
Will handle this in next Innovation day. Target with other CI changes to Onnxrt since we're trying to reduce workloads through CI right now.
Onnxrt seems to be a blocker when using vega nodes.
Need to identify when code changes break parity changes with onnxruntime's code changes for their parity operations which are used in DeepLearning Models
Add a CI stage as part of onnxruntime run in parallel to CI unit tests, to determine if MIGraphX changes break DLM related checks/tests
We're already building onnxruntime we just need to add support to build the python wheel as well as support for the current libraries we're pulling that DLM uses.
This should give us a heads up between
test_gelu_parity --no_optimize
test_parity_layernorm --no_optimize
test_parity_huggingface_gpt_attention.py
[ ] Move control of libraries into MIGraphX repo that are required to run parity tests into CI
[ ] Split out unit testing in build_and_test.sh
[ ] Build onnxruntime wheel as part of onnxruntime build for install
[ ] Add CI stage in parallel for parity tests
[ ] Mirror what DLM is doing without tracking perf (just look at pass/fail)