[fmha] support head dim 512 #1184
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: run unit tests on Intel GPU. | |
| on: | |
| pull_request: | |
| branches: [main] | |
| push: | |
| branches: [main] | |
| permissions: | |
| contents: read | |
| env: | |
| REGISTRY: localhost:5000 | |
| jobs: | |
| build-docker-image-latest-pvc: | |
| runs-on: self-hosted-pvc | |
| steps: | |
| - name: Clean workspace | |
| run: | | |
| sudo chown -R "$(id -u):$(id -g)" . || true | |
| git clean -ffdx || true | |
| git reset --hard || true | |
| - name: Checkout | |
| uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1 | |
| with: | |
| set-safe-directory: true | |
| clean: true | |
| - name: build docker image & push to local | |
| id: build-image | |
| run: | | |
| docker build -t xpu-kernel-ci-image:latest -f Dockerfile.xpu . | |
| docker tag xpu-kernel-ci-image:latest ${{ env.REGISTRY }}/xpu-kernel-ci-image:latest | |
| docker push ${{ env.REGISTRY }}/xpu-kernel-ci-image:latest | |
| build-docker-image-latest-bmg: | |
| runs-on: self-hosted-bmg | |
| steps: | |
| - name: Clean workspace | |
| run: | | |
| sudo chown -R "$(id -u):$(id -g)" . || true | |
| git clean -ffdx || true | |
| git reset --hard || true | |
| - name: Checkout | |
| uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1 | |
| with: | |
| set-safe-directory: true | |
| clean: true | |
| - name: build docker image & push to local | |
| id: build-image | |
| run: | | |
| docker build -t xpu-kernel-ci-image:latest -f Dockerfile.xpu . | |
| docker tag xpu-kernel-ci-image:latest ${{ env.REGISTRY }}/xpu-kernel-ci-image:latest | |
| docker push ${{ env.REGISTRY }}/xpu-kernel-ci-image:latest | |
| run-unit-tests-pvc: | |
| runs-on: self-hosted-pvc | |
| needs: build-docker-image-latest-pvc | |
| timeout-minutes: 40 | |
| container: | |
| image: localhost:5000/xpu-kernel-ci-image:latest | |
| options: --device /dev/dri -v /dev/dri/by-path:/dev/dri/by-path --privileged -v ccache:/root/.ccache -e CCACHE_DIR=/root/.ccache | |
| steps: | |
| - name: Checkout | |
| uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1 | |
| with: | |
| set-safe-directory: true | |
| - name: build & install wheel | |
| run: | | |
| git config --global --add safe.directory "${GITHUB_WORKSPACE}" | |
| export CCACHE_DIR=/root/.ccache | |
| export CCACHE_BASEDIR="${GITHUB_WORKSPACE}" | |
| export CCACHE_NOHASHDIR=1 | |
| export CCACHE_COMPILERCHECK=content | |
| ccache -s || true | |
| ccache -p || true | |
| git submodule sync && git submodule update --init --recursive | |
| uv pip install -r requirements.txt | |
| MAX_JOBS=128 uv pip install --no-build-isolation -e . -v | |
| ccache -s || true | |
| - name: test | |
| run: | | |
| ZE_AFFINITY_MASK=0,1 SKIP_HANG_KERNEL=1 SKIP_ACC_ERROR_KERNEL=1 pytest -v -s tests/ | |
| VLLM_XPU_FORCE_XE_DEFAULT_KERNEL=1 ZE_AFFINITY_MASK=0,1 pytest -v -s tests/fused_moe/test_grouped_gemm.py::test_grouped_gemm | |
| clean-repo-pvc: | |
| runs-on: self-hosted-pvc | |
| needs: run-unit-tests-pvc | |
| steps: | |
| - name: Clean workspace | |
| run: | | |
| sudo chown -R "$(id -u):$(id -g)" . || true | |
| git clean -ffdx || true | |
| git reset --hard || true | |
| run-unit-tests-bmg: | |
| runs-on: self-hosted-bmg | |
| needs: build-docker-image-latest-bmg | |
| timeout-minutes: 40 | |
| container: | |
| image: localhost:5000/xpu-kernel-ci-image:latest | |
| options: --device /dev/dri -v /dev/dri/by-path:/dev/dri/by-path --privileged -v ccache:/root/.ccache -e CCACHE_DIR=/root/.ccache | |
| steps: | |
| - name: Checkout | |
| uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1 | |
| with: | |
| set-safe-directory: true | |
| - name: build & install wheel | |
| run: | | |
| git config --global --add safe.directory "${GITHUB_WORKSPACE}" | |
| export CCACHE_DIR=/root/.ccache | |
| export CCACHE_BASEDIR="${GITHUB_WORKSPACE}" | |
| export CCACHE_NOHASHDIR=1 | |
| export CCACHE_COMPILERCHECK=content | |
| ccache -s || true | |
| ccache -p || true | |
| git submodule sync && git submodule update --init --recursive | |
| uv pip install -r requirements.txt | |
| MAX_JOBS=80 uv pip install --no-build-isolation -e . -v | |
| ccache -s || true | |
| - name: test | |
| run: | | |
| # tests/test_moe_align_block_size.py, tests/test_moe_lora_align_sum.py takes much time than expected. ignore it for now. | |
| ZE_AFFINITY_MASK=0,1 pytest -v -s tests/ --ignore=tests/test_lora_ops.py --ignore=tests/test_fp8_quant.py --ignore=tests/test_moe_align_block_size.py --ignore=tests/test_moe_lora_align_sum.py --ignore=tests/test_cache.py::test_swap_blocks --ignore=tests/test_topk_per_row.py --ignore=tests/test_lora_ops.py | |
| # fixme: Running lora UT separately to avoid OOM when running together with other tests. | |
| ZE_AFFINITY_MASK=0,1 pytest -v -s tests/test_lora_ops.py | |
| VLLM_XPU_FORCE_XE_DEFAULT_KERNEL=1 ZE_AFFINITY_MASK=0,1 pytest -v -s tests/fused_moe/test_grouped_gemm.py::test_grouped_gemm | |
| clean-repo-bmg: | |
| runs-on: self-hosted-bmg | |
| needs: run-unit-tests-bmg | |
| steps: | |
| - name: Clean workspace | |
| run: | | |
| sudo chown -R "$(id -u):$(id -g)" . || true | |
| git clean -ffdx || true | |
| git reset --hard || true |