|
47 | 47 | RUN git checkout main
|
48 | 48 |
|
49 | 49 | # Pinning versions in requirements might be good practice for CI consistency
|
50 |
| - RUN pip install pytest pytest_asyncio |
| 50 | + RUN pip install pytest pytest_asyncio nixl==0.4.1 |
51 | 51 | RUN pip install git+https://github.com/EleutherAI/lm-evaluation-harness.git
|
| 52 | + RUN pip install lm-eval[api] |
52 | 53 |
|
53 | 54 | ENV no_proxy=localhost,127.0.0.1
|
54 | 55 | ENV PT_HPU_ENABLE_LAZY_COLLECTIVES=true
|
|
94 | 95 |
|
95 | 96 | EXITCODE=$?
|
96 | 97 | echo "Test script exited with code: $EXITCODE"
|
| 98 | + hpu_pd_tests: |
| 99 | + runs-on: ucb-vllm-cicd-g2 |
| 100 | + # This is a final job that runs after the build and unit tests |
| 101 | + # Unit tests are supposed to fail fast if anything goes wrong, removing the need for this job to run. |
| 102 | + needs: [pre_merge_hpu_test_build, hpu_unit_tests] |
| 103 | + steps: |
| 104 | + - name: Run test scripts |
| 105 | + run: | |
| 106 | + EXITCODE=1 |
| 107 | + remove_docker_containers() { docker rm -f hpu-plugin-v1-test-pd-tests-${{ github.event.pull_request.head.sha }} || true; } |
| 108 | + trap 'remove_docker_containers; exit $EXITCODE;' EXIT |
| 109 | + remove_docker_containers |
| 110 | +
|
| 111 | + echo "Running HPU plugin v1 nixl pd tests" |
| 112 | + docker run --rm --runtime=habana --name=hpu-plugin-v1-test-pd-tests-${{ github.event.pull_request.head.sha }} --network=host \ |
| 113 | + -e HABANA_VISIBLE_DEVICES=all \ |
| 114 | + -e HF_HOME=/workspace/hf_cache \ |
| 115 | + -v /mnt/hf_cache:/workspace/hf_cache \ |
| 116 | + hpu-plugin-v1-test-env-pre-merge-${{ github.event.pull_request.head.sha }} \ |
| 117 | + /bin/bash "/workspace/vllm-gaudi/tests/v1/kv_connector/nixl_integration/run_accuracy_test.sh" |
97 | 118 |
|
| 119 | + EXITCODE=$? |
| 120 | + echo "Test script exited with code: $EXITCODE" |
98 | 121 | hpu_e2e_tests:
|
99 | 122 | runs-on: ucb-vllm-cicd-g2
|
100 | 123 | # This is a final job that runs after the build and unit tests
|
|
0 commit comments