Update on "[MoE][2/n]Move EP setup from trainer to config registry an… #2701
Workflow file for this run
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: GraphTrainer 8 GPU Integration Tests | |
| on: | |
| push: | |
| branches: [ main ] | |
| tags: | |
| - ciflow/8gpu/* | |
| paths: | |
| - 'torchtitan/experiments/graph_trainer/**' | |
| - '.github/workflows/integration_test_8gpu_graph_trainer.yaml' | |
| pull_request: | |
| types: [opened, synchronize, reopened, ready_for_review] | |
| paths: | |
| - 'torchtitan/experiments/graph_trainer/**' | |
| - '.github/workflows/integration_test_8gpu_graph_trainer.yaml' | |
| schedule: | |
| # Runs every 12 hours | |
| - cron: '0 */12 * * *' | |
| concurrency: | |
| group: unit-test${{ github.workflow }}-${{ github.ref == 'refs/heads/main' && github.run_number || github.ref }} | |
| cancel-in-progress: true | |
| defaults: | |
| run: | |
| shell: bash -l -eo pipefail {0} | |
| permissions: | |
| id-token: write | |
| contents: read | |
| jobs: | |
| build-test: | |
| if: github.event_name != 'pull_request' || github.event.pull_request.draft == false | |
| uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main | |
| with: | |
| runner: linux.g5.48xlarge.nvidia.gpu | |
| gpu-arch-type: cuda | |
| gpu-arch-version: "12.8" | |
| docker-image: torchtitan-ubuntu-20.04-clang12 | |
| repository: pytorch/torchtitan | |
| upload-artifact: outputs | |
| timeout: 45 | |
| script: | | |
| set -eux | |
| # The generic Linux job chooses to use base env, not the one setup by the image | |
| CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]") | |
| conda activate "${CONDA_ENV}" | |
| # Log GPU info / driver version for debugging. | |
| DRIVER_VERSION=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader | head -n 1 || true) | |
| echo "CUDA driver version: ${DRIVER_VERSION}" | |
| pip config --user set global.progress_bar off | |
| python -m pip install --force-reinstall --pre \ | |
| torch --index-url https://download.pytorch.org/whl/nightly/cu128 | |
| python -m pip install torchdata==0.12.0.dev20260327 --extra-index-url https://download.pytorch.org/whl/nightly/cpu | |
| sudo mkdir -p "$RUNNER_TEMP/artifacts-to-be-uploaded" | |
| sudo chown -R $(id -u):$(id -g) "$RUNNER_TEMP/artifacts-to-be-uploaded" | |
| python -m torchtitan.experiments.graph_trainer.tests.integration_tests --test_suite graph_trainer_default --gpu_arch_type cuda $RUNNER_TEMP/artifacts-to-be-uploaded --ngpu 8 | |
| # Run the numerics unit tests (dense models only; MoE tests run in the H100 workflow) | |
| pytest torchtitan/experiments/graph_trainer/tests/test_numerics.py::TestSimpleFSDP -v | |
| pytest torchtitan/experiments/graph_trainer/tests/test_numerics.py::TestGraphTrainerNumerics -v -k "dense" | |
| # Run the graph passes unit tests | |
| torchrun --nproc-per-node=8 -m pytest torchtitan/experiments/graph_trainer/tests/test_passes.py::TestReassignToPgPass -v | |
| pytest torchtitan/experiments/graph_trainer/tests/test_passes.py::TestApplySACPass -v | |
| # Run the make_fx tracer unit tests | |
| pytest torchtitan/experiments/graph_trainer/tests/test_trace_module.py -v -k "TestTraceModule or TestTraceDTensor or TestMetadataPropagation" | |
| # Run precompile unit tests | |
| pytest torchtitan/experiments/graph_trainer/tests/test_precompile.py -v | |
| # Run bitwise deterministic guardrail test | |
| pytest torchtitan/experiments/graph_trainer/tests/test_bitwise_deterministic.py -v | |
| rm -rf $RUNNER_TEMP/artifacts-to-be-uploaded/*/checkpoint |