vllm-ascend 0.13.0rc1__tar.gz → 0.13.0rc2__tar.gz

vllm_ascend-0.13.0rc2/.github/Dockerfile.buildwheel

@@ -0,0 +1,45 @@
+#
+# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# This file is a part of the vllm-ascend project.
+#
+ARG PY_VERSION=3.11
+FROM quay.io/ascend/manylinux:8.5.0-910b-manylinux_2_28-py${PY_VERSION}
+
+ARG SOC_VERSION="ascend910b1"
+
+# Define environments
+ENV DEBIAN_FRONTEND=noninteractive
+ENV SOC_VERSION=$SOC_VERSION
+RUN yum update -y && \
+    yum install -y python3-pip git vim wget net-tools gcc gcc-c++ make cmake numactl-devel && \
+    rm -rf /var/cache/yum
+
+WORKDIR /workspace
+
+COPY . /workspace/vllm-ascend/
+
+# Install req
+RUN python3 -m pip install -r vllm-ascend/requirements.txt --extra-index https://download.pytorch.org/whl/cpu/ && \
+    python3 -m pip install twine
+
+# Install vllm-ascend
+RUN source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
+    source /usr/local/Ascend/nnal/atb/set_env.sh && \
+    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
+    cd vllm-ascend && \
+    python3 setup.py bdist_wheel && \
+    ls -l dist 
+
+CMD ["/bin/bash"]

vllm_ascend-0.13.0rc2/.github/actionlint.yaml

@@ -0,0 +1,23 @@
+self-hosted-runner:
+  # Labels of self-hosted runner in array of strings.
+  labels:
+    - linux-aarch64-a2-0
+    - linux-aarch64-a2-1
+    - linux-aarch64-a2-2
+    - linux-aarch64-a2-4
+    - linux-aarch64-a2-8
+    - linux-arm64-npu-static-8
+    - linux-aarch64-310p-1
+    - linux-aarch64-310p-2
+    - linux-aarch64-310p-4
+    - ubuntu-24.04-arm
+    - linux-aarch64-a3-1
+    - linux-aarch64-a3-2
+    - linux-aarch64-a3-4
+    - linux-aarch64-a3-8
+    - linux-amd64-cpu-0
+    - linux-amd64-cpu-8
+    - linux-amd64-cpu-16
+    - linux-aarch64-a3-0
+    - linux-amd64-cpu-8-hk
+    - linux-amd64-cpu-16-hk

vllm_ascend-0.13.0rc2/.github/workflows/_e2e_nightly_multi_node.yaml

@@ -0,0 +1,299 @@
+name: 'e2e nightly test multi_node'
+
+on:
+  workflow_call:
+    inputs:
+      soc_version:
+        required: true
+        type: string
+        description: use a2 or a3
+      runner:
+        required: false
+        type: string
+        default: linux-aarch64-a3-0
+      image:
+        required: false
+        type: string
+        description: base image for pods
+        default: "swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-910b-ubuntu22.04-py3.11"
+      config_file_path:
+        required: true
+        type: string
+        description: the model config for multi_node test
+      replicas:
+        required: false
+        default: "1"
+        type: string
+        description: replicas of the k8s cluster
+      size:
+        required: false
+        default: "2"
+        type: string
+        description: how many pods will be pulled up via lws.yaml, indicates number of nodes we need
+      vllm_version:
+        required: false
+        default: "v0.13.0"
+        type: string
+        description: vllm version to use
+      vllm_ascend_remote_url:
+        required: false
+        default: https://github.com/vllm-project/vllm-ascend.git
+        type: string
+        description: used for pr level tests
+      vllm_ascend_ref:
+        required: false
+        default: main
+        type: string
+        description: used for pr level tests
+    secrets:
+      KUBECONFIG_B64:
+        required: true
+
+
+# Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
+# declared as "shell: bash -el {0}" on steps that need to be properly activated.
+# It's used to activate ascend-toolkit environment variables.
+defaults:
+  run:
+    shell: bash -el {0}
+
+# only cancel in-progress runs of the same workflow
+# and ignore the lint / 8 cards test type
+concurrency:
+  group: ascend-nightly-${{ github.workflow_ref }}-${{ github.ref }}-${{ inputs.soc_version }}
+  cancel-in-progress: true
+
+jobs:
+  e2e:
+    name: ${{ inputs.config_file_path }}
+    # This is the runner with no NPU for k8s controller
+    runs-on: ${{ inputs.runner }}
+    container:
+      image: m.daocloud.io/quay.io/ascend/cann:8.5.0-a3-ubuntu22.04-py3.11
+      env:
+        KUBECONFIG: /tmp/kubeconfig
+        KUBECTL: /root/.cache/.kube/kubectl
+        NAMESPACE: vllm-project
+        LEADER_POD: vllm-0
+        RESULT_FILE: /root/.cache/tests/ret_${{ inputs.soc_version }}
+    steps:
+        - name: Install system denpendencies
+          run: |
+           # configure apt and pip source
+           sed -i 's|ports.ubuntu.com|mirrors.tuna.tsinghua.edu.cn|g' /etc/apt/sources.list
+           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
+           pip install jinja2-cli
+
+        - name: Install kubectl
+          run: |
+            # Install kubectl
+            arch=$(uname -m)
+
+            if echo "$arch" | grep -qiE "arm|aarch64"; then
+              echo "Detected ARM architecture: $arch"
+              KUBECTL="$KUBECTL"_arm
+            fi
+            install -o root -g root -m 0755 $KUBECTL /usr/local/bin/kubectl
+
+            # Verify kubectl installation
+            kubectl version --client=true
+
+        - name: Decode kubeconfig from secrets
+          run: |
+            # Decode and save kubeconfig
+            echo "${{ secrets.KUBECONFIG_B64 }}" | base64 -d > $KUBECONFIG
+
+        - name: Checkout code
+          uses: actions/checkout@v6
+
+        - name: Prepare scripts
+          run: |
+            # prepare for lws entrypoint scripts
+            install -D tests/e2e/nightly/multi_node/scripts/run.sh /root/.cache/tests/run.sh
+            # clear log directory
+            rm -fr $RESULT_FILE
+
+        - name: Clear resources
+          run: |
+            set -euo pipefail
+
+            CRD_NAME="${CRD_NAME:-vllm}"
+            TIMEOUT=${TIMEOUT:-120}
+            SLEEP_INTERVAL=2
+
+            echo "Deleting leaderworkerset [$CRD_NAME] in namespace [$NAMESPACE]..."
+            kubectl delete leaderworkerset "$CRD_NAME" -n "$NAMESPACE" --ignore-not-found
+
+            echo "Waiting for all pods starting with 'vllm' to be deleted..."
+            START_TIME=$(date +%s)
+
+            while true; do
+              NOW=$(date +%s)
+              ELAPSED=$((NOW - START_TIME))
+
+              if [[ $ELAPSED -ge $TIMEOUT ]]; then
+                echo "Timeout reached ($TIMEOUT seconds), some pods still exist:"
+                kubectl get pods -n "$NAMESPACE" | grep '^vllm' || true
+                exit 1
+              fi
+
+              PODS_EXIST=$(kubectl get pods -n "$NAMESPACE" -o jsonpath='{.items[*].metadata.name}' 2>/dev/null | tr ' ' '\n' | grep '^vllm' || true)
+
+              if [[ -z "$PODS_EXIST" ]]; then
+                echo "All vllm pods deleted."
+                break
+              else
+                echo "Waiting for pods to be deleted: $PODS_EXIST"
+                sleep $SLEEP_INTERVAL
+              fi
+            done
+
+        - name: Launch cluster
+          id: launcher
+          run: |
+            set -e
+
+            size="${{ inputs.size }}"
+            replicas="${{ inputs.replicas }}"
+            image="${{ inputs.image }}"
+            config_file_path="${{ inputs.config_file_path }}"
+            vllm_version="${{ inputs.vllm_version }}"
+            vllm_ascend_ref="${{ inputs.vllm_ascend_ref }}"
+            vllm_ascend_remote_url="${{ inputs.vllm_ascend_remote_url }}"
+            result_file_path="$RESULT_FILE"
+            fail_tag=FAIL_TAG_"${{ inputs.config_file_path }}"
+            echo "FAIL_TAG=${fail_tag}" >> $GITHUB_ENV
+
+            required_params=("size" "replicas" "image" "config_file_path")
+            for param in "${required_params[@]}"; do
+              if [ -z "${!param}" ]; then
+                echo "Error: Parameter '$param' is required but empty"
+                exit 1
+              fi
+            done
+
+            if [ "${{ inputs.soc_version }}" = "a3" ]; then
+              npu_per_node=16
+            else
+              npu_per_node=8
+            fi
+
+            jinja2 tests/e2e/nightly/multi_node/scripts/lws.yaml.jinja2 \
+              -D size="$size" \
+              -D replicas="$replicas" \
+              -D image="$image" \
+              -D config_file_path="$config_file_path" \
+              -D vllm_version="$vllm_version" \
+              -D vllm_ascend_remote_url="$vllm_ascend_remote_url" \
+              -D vllm_ascend_ref="$vllm_ascend_ref" \
+              -D result_file_path="$result_file_path" \
+              -D npu_per_node="$npu_per_node" \
+              -D fail_tag="$fail_tag" \
+              --outfile lws.yaml
+
+            kubectl apply -f ./lws.yaml
+
+        - name: Waiting for pod ready
+          run: |
+            POD_PREFIX="${POD_PREFIX:-vllm-0}"
+            SIZE="${{ inputs.size }}"
+            TIMEOUT=1200  # default timeout 20 minutes
+
+            echo "Waiting for Pods in namespace [$NAMESPACE] to become Running and Ready (timeout ${TIMEOUT}s)..."
+
+            START_TIME=$(date +%s)
+
+            while true; do
+              NOW=$(date +%s)
+              ELAPSED=$((NOW - START_TIME))
+              if [[ $ELAPSED -ge $TIMEOUT ]]; then
+                echo "Timeout reached after ${ELAPSED}s"
+                echo "Dumping pod status for debugging:"
+                kubectl get pods -n "$NAMESPACE"
+                kubectl describe pod "$LEADER_POD" -n "$NAMESPACE"
+                exit 1
+              fi
+
+              # 1) check follower pods
+              ALL_FOLLOWERS_READY=true
+              for ((i=1; i<SIZE; i++)); do
+                POD="${POD_PREFIX}-${i}"
+                PHASE=$(kubectl get pod "$POD" -n "$NAMESPACE" -o jsonpath='{.status.phase}' 2>/dev/null || echo "NotFound")
+                READY=$(kubectl get pod "$POD" -n "$NAMESPACE" -o jsonpath='{.status.containerStatuses[*].ready}' 2>/dev/null)
+
+                echo "Follower [$POD] phase=$PHASE ready=$READY"
+
+                if [[ "$PHASE" != "Running" || "$READY" != "true" ]]; then
+                  echo "Follower [$POD] not Ready yet..."
+                  ALL_FOLLOWERS_READY=false
+                  break
+                fi
+              done
+
+              # 2) check leader pod
+              LEADER_PHASE=$(kubectl get pod "$LEADER_POD" -n "$NAMESPACE" -o jsonpath='{.status.phase}' 2>/dev/null || echo "NotFound")
+              LEADER_READY=$(kubectl get pod "$LEADER_POD" -n "$NAMESPACE" -o jsonpath='{.status.containerStatuses[*].ready}' 2>/dev/null)
+
+              echo "Leader [$LEADER_POD] phase=$LEADER_PHASE ready=$LEADER_READY"
+
+              if [[ "$LEADER_PHASE" != "Running" || "$LEADER_READY" != "true" ]]; then
+                echo "Leader not Ready yet..."
+                ALL_FOLLOWERS_READY=false
+              fi
+
+              if [[ "$ALL_FOLLOWERS_READY" == "true" ]]; then
+                echo "All follower pods and leader pod are Running and Ready — continuing."
+                break
+              fi
+
+              sleep 2
+            done
+
+        - name: Stream logs
+          run: |
+            set -euo pipefail
+
+            size="${{ inputs.size }}"
+            pids=()
+
+            cleanup() {
+              echo "Cleaning up background log streams..."
+              for pid in "${pids[@]}"; do
+                kill "$pid" 2>/dev/null || true
+              done
+            }
+            trap cleanup EXIT
+
+            for i in $(seq 1 $((size - 1))); do
+              POD="vllm-0-${i}"
+
+              echo "==== Collecting logs from worker pod: $POD ===="
+              kubectl logs -f "$POD" -n "$NAMESPACE" \
+                > "/tmp/${POD}_logs.txt" 2>&1 &
+
+              pids+=($!)
+            done
+
+            echo "==== Streaming logs from leader pod: $LEADER_POD ===="
+            echo "Looking for logs containing: $FAIL_TAG"
+
+            kubectl logs -f "$LEADER_POD" -n "$NAMESPACE" | while IFS= read -r line; do
+              echo "$line"
+              if echo "$line" | grep -q "$FAIL_TAG"; then
+                exit 1
+              fi
+            done
+
+        - name: Upload logs
+          if: always()
+          uses: actions/upload-artifact@v6
+          with:
+            name: ${{ inputs.config_file_path }}-pod-logs
+            path: /tmp/vllm*_logs.txt
+            retention-days: 7
+
+        - name: Post process
+          if: always()
+          run: |
+            kubectl get pods -n $NAMESPACE --ignore-not-found=true
+            kubectl delete -f ./lws.yaml --ignore-not-found=true || true

vllm_ascend-0.13.0rc2/.github/workflows/_e2e_nightly_single_node.yaml

@@ -0,0 +1,138 @@
+#
+# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# This file is a part of the vllm-ascend project.
+#
+
+name: 'e2e nightly test'
+
+on:
+  workflow_call:
+    inputs:
+      vllm:
+        required: true
+        type: string
+      runner:
+        required: true
+        type: string
+      image:
+        required: false
+        type: string
+        default: "swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-910b-ubuntu22.04-py3.11"
+      tests:
+        required: true
+        type: string
+      name:
+        required: false
+        type: string
+
+# Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
+# declared as "shell: bash -el {0}" on steps that need to be properly activated.
+# It's used to activate ascend-toolkit environment variables.
+defaults:
+  run:
+    shell: bash -el {0}
+
+# only cancel in-progress runs of the same workflow
+# and ignore the lint / 1 card / 4 cards test type
+concurrency:
+  group: ascend-nightly-${{ github.workflow_ref }}-${{ github.ref }}-${{ inputs.tests }}
+  cancel-in-progress: true
+
+jobs:
+  e2e-nightly:
+    name: ${{ inputs.tests }}
+    runs-on: ${{ inputs.runner }}
+    timeout-minutes: 600
+    container:
+      image: ${{ inputs.image }}
+    env:
+      TRANSFORMERS_OFFLINE: 1
+      VLLM_USE_MODELSCOPE: True
+    steps:
+      - name: Check npu and CANN info
+        run: |
+          npu-smi info
+          cat /usr/local/Ascend/ascend-toolkit/latest/"$(uname -i)"-linux/ascend_toolkit_install.info
+
+      - name: Show vLLM and vLLM-Ascend version
+        working-directory: /vllm-workspace
+        run: |
+          echo "Installed vLLM-related Python packages:"
+          pip list | grep vllm || echo "No vllm packages found."
+
+          echo ""
+          echo "============================"
+          echo "vLLM Git information"
+          echo "============================"
+          cd vllm
+          if [ -d .git ]; then
+            echo "Branch:      $(git rev-parse --abbrev-ref HEAD)"
+            echo "Commit hash: $(git rev-parse HEAD)"
+            echo "Author:      $(git log -1 --pretty=format:'%an <%ae>')"
+            echo "Date:        $(git log -1 --pretty=format:'%ad' --date=iso)"
+            echo "Message:     $(git log -1 --pretty=format:'%s')"
+            echo "Tags:        $(git tag --points-at HEAD || echo 'None')"
+            echo "Remote:      $(git remote -v | head -n1)"
+            echo ""
+          else
+            echo "No .git directory found in vllm"
+          fi
+          cd ..
+
+          echo ""
+          echo "============================"
+          echo "vLLM-Ascend Git information"
+          echo "============================"
+          cd vllm-ascend
+          if [ -d .git ]; then
+            echo "Branch:      $(git rev-parse --abbrev-ref HEAD)"
+            echo "Commit hash: $(git rev-parse HEAD)"
+            echo "Author:      $(git log -1 --pretty=format:'%an <%ae>')"
+            echo "Date:        $(git log -1 --pretty=format:'%ad' --date=iso)"
+            echo "Message:     $(git log -1 --pretty=format:'%s')"
+            echo "Tags:        $(git tag --points-at HEAD || echo 'None')"
+            echo "Remote:      $(git remote -v | head -n1)"
+            echo ""
+          else
+            echo "No .git directory found in vllm-ascend"
+          fi
+          cd ..
+
+      - name: Install custom-ops (for DeepSeek-V3.2-Exp)
+        if: ${{ inputs.name == 'deepseek3_2-exp-w8a8' }}
+        shell: bash -l {0}
+        run: |
+          wget https://vllm-ascend.obs.cn-north-4.myhuaweicloud.com/vllm-ascend/a3/CANN-custom_ops-sfa-linux.aarch64.run
+          chmod +x ./CANN-custom_ops-sfa-linux.aarch64.run
+          ./CANN-custom_ops-sfa-linux.aarch64.run --quiet
+          export ASCEND_CUSTOM_OPP_PATH=/usr/local/Ascend/ascend-toolkit/latest/opp/vendors/customize:${ASCEND_CUSTOM_OPP_PATH}
+          export LD_LIBRARY_PATH=/usr/local/Ascend/ascend-toolkit/latest/opp/vendors/customize/op_api/lib/:${LD_LIBRARY_PATH}
+          wget https://vllm-ascend.obs.cn-north-4.myhuaweicloud.com/vllm-ascend/a3/custom_ops-1.0-cp311-cp311-linux_aarch64.whl
+          pip install custom_ops-1.0-cp311-cp311-linux_aarch64.whl
+          . /usr/local/Ascend/ascend-toolkit/set_env.sh
+
+      - name: Run vllm-project/vllm-ascend test
+        env:
+          VLLM_WORKER_MULTIPROC_METHOD: spawn
+          VLLM_USE_MODELSCOPE: True
+          VLLM_CI_RUNNER: ${{ inputs.runner }}
+          BENCHMARK_HOME: /vllm-workspace/vllm-ascend/benchmark
+        working-directory: /vllm-workspace/vllm-ascend
+        run: |
+          # ignore test_dispatch_ffn_combine until the test is fixed
+          pytest -sv ${{ inputs.tests }} \
+          --ignore=tests/e2e/nightly/single_node/ops/singlecard_ops/test_fused_moe.py
+
+