|
| 1 | +name: ~NGC release testing |
| 2 | + |
| 3 | +on: |
| 4 | + workflow_dispatch: |
| 5 | + inputs: |
| 6 | + JAX_IMAGE: |
| 7 | + type: string |
| 8 | + description: "JAX image to run tests on" |
| 9 | + required: false |
| 10 | + default: '' |
| 11 | + MAXTEXT_IMAGE: |
| 12 | + type: string |
| 13 | + description: "MaxText image to run tests on" |
| 14 | + required: false |
| 15 | + default: '' |
| 16 | + |
| 17 | + |
| 18 | +concurrency: |
| 19 | + group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }} |
| 20 | + |
| 21 | +permissions: |
| 22 | + contents: read # to fetch code |
| 23 | + actions: write # to cancel previous workflows |
| 24 | + packages: write # to upload container |
| 25 | + |
| 26 | +jobs: |
| 27 | + test-nccl: |
| 28 | + if: inputs.JAX_IMAGE != '' |
| 29 | + uses: ./.github/workflows/_test_nccl.yaml |
| 30 | + with: |
| 31 | + CONTAINER: ${{ inputs.JAX_IMAGE }} |
| 32 | + secrets: inherit |
| 33 | + |
| 34 | + test-jax: |
| 35 | + if: inputs.JAX_IMAGE != '' |
| 36 | + uses: ./.github/workflows/_test_unit.yaml |
| 37 | + with: |
| 38 | + TEST_NAME: jax |
| 39 | + EXECUTE: | |
| 40 | + docker run -i --shm-size=1g --gpus all \ |
| 41 | + ${{ inputs.JAX_IMAGE }} \ |
| 42 | + bash <<"EOF" |& tee test-backend-independent.log |
| 43 | + test-jax.sh -b backend-independent |
| 44 | + EOF |
| 45 | + docker run -i --shm-size=1g --gpus all \ |
| 46 | + ${{ inputs.JAX_IMAGE }} \ |
| 47 | + bash <<"EOF" |& tee tee test-gpu.log |
| 48 | + test-jax.sh -b gpu |
| 49 | + EOF |
| 50 | + STATISTICS_SCRIPT: | |
| 51 | + errors=$(cat test-*.log | grep -c 'ERROR:' || true) |
| 52 | + failed_tests=$(cat test-*.log | grep -c 'FAILED in' || true) |
| 53 | + passed_tests=$(cat test-*.log | grep -c 'PASSED in' || true) |
| 54 | + total_tests=$((failed_tests + passed_tests)) |
| 55 | + echo "TOTAL_TESTS=${total_tests}" >> $GITHUB_OUTPUT |
| 56 | + echo "ERRORS=${errors}" >> $GITHUB_OUTPUT |
| 57 | + echo "PASSED_TESTS=${passed_tests}" >> $GITHUB_OUTPUT |
| 58 | + echo "FAILED_TESTS=${failed_tests}" >> $GITHUB_OUTPUT |
| 59 | + ARTIFACTS: | |
| 60 | + test-backend-independent.log |
| 61 | + test-gpu.log |
| 62 | + secrets: inherit |
| 63 | + |
| 64 | + test-maxtext: |
| 65 | + if: inputs.MAXTEXT_IMAGE != '' |
| 66 | + uses: ./.github/workflows/_test_maxtext.yaml |
| 67 | + with: |
| 68 | + MAXTEXT_IMAGE: ${{ inputs.MAXTEXT_IMAGE }} |
| 69 | + secrets: inherit |
| 70 | + |
| 71 | + test-maxtext-eks: |
| 72 | + if: inputs.MAXTEXT_IMAGE != '' |
| 73 | + uses: ./.github/workflows/_test_maxtext_k8s.yaml |
| 74 | + with: |
| 75 | + MAXTEXT_IMAGE: ${{ inputs.MAXTEXT_IMAGE }} |
| 76 | + secrets: inherit |
| 77 | + |
| 78 | + finalize: |
| 79 | + needs: [ test-nccl, test-jax, test-maxtext, test-maxtext-eks ] |
| 80 | + if: "!cancelled()" |
| 81 | + uses: ./.github/workflows/_finalize.yaml |
| 82 | + secrets: inherit |
0 commit comments