diff --git a/.github/workflows/_npu-pr-test-stage.yml b/.github/workflows/_npu-pr-test-stage.yml
new file mode 100644
index 000000000..db1171fa9
--- /dev/null
+++ b/.github/workflows/_npu-pr-test-stage.yml
@@ -0,0 +1,213 @@
+name: PR Test Stage for NPU
+# Reusable workflow for one CUDA test stage. Caller pr-test-npu.yml forwards
+
+on:
+  workflow_call:
+    inputs:
+      self_name:
+        description: 'Caller job key; used for partitions[suite] lookup.'
+        type: string
+        required: true
+      runner_config:
+        description: 'Key in scripts/ci/runner_configs.yml. Resolves install script, artifact version, install timeout, runs-on label, and rdma_devices.'
+        type: string
+        required: true
+      image:
+        description: 'The container image for NPU test environment.'
+        type: string
+        required: true
+      npu_device_type:
+        description: 'NPU device type, e.g. 910b, a3.'
+        type: string
+        default: 'a3'
+      run_timeout_minutes:
+        description: 'timeout-minutes for the Run test step. Required so compute_partitions.py can read it from pr-test.yml without a duplicated default constant.'
+        type: string
+        required: true
+      timeout_per_file:
+        description: 'run_suite.py --timeout-per-file value (empty = unset).'
+        type: string
+        default: ''
+      partitions:
+        description: 'partitions config, e.g. {"size":1,"arr":[0]}'
+        type: string
+        default: '{"size":1,"arr":[0]}'
+      ref:
+        description: 'Git ref (branch, tag, or SHA) to test. If not provided, uses the default branch.'
+        type: string
+        default: ''
+      skip_pr_test_health_check:
+        description: 'Git ref (branch, tag, or SHA) to test. If not provided, uses the default branch.'
+        type: string
+        default: 'false'
+      github-token:
+        description: 'GitHub token for API calls'
+        type: string
+        default: ${{ github.token }}
+
+env:
+  SKIP_PR_TEST_HEALTH_CHECK: ${{ inputs.skip_pr_test_health_check }}
+  SGLANG_USE_MODELSCOPE: true
+  SGLANG_IS_IN_CI: true
+  HF_ENDPOINT: https://hf-mirror.com
+  TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
+  PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
+  STREAMS_PER_DEVICE: 32
+
+jobs:
+  run:
+    runs-on: ${{ inputs.runner_config }}
+    container:
+      image: ${{ inputs.image }}
+    timeout-minutes: 240
+    strategy:
+      fail-fast: false
+      matrix:
+        partition: ${{ fromJson(inputs.partitions).arr }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Mark repository safe
+        run: |
+          git config --system --add safe.directory ${GITHUB_WORKSPACE}
+
+      - name: Check PR test health
+        uses: actions/github-script@v8
+        env:
+          SKIP_PR_TEST_HEALTH_CHECK: ${{ env.SKIP_PR_TEST_HEALTH_CHECK }}
+        with:
+          github-token: ${{ inputs.github-token || github.token }}
+          script: |
+            core.notice(`[health-check] START — event=${context.eventName}, runId=${context.runId}`);
+
+            // Skip when explicitly requested via env var (e.g. release branch cut)
+            if (process.env.SKIP_PR_TEST_HEALTH_CHECK === 'true') {
+              core.notice('[health-check] SKIP: SKIP_PR_TEST_HEALTH_CHECK=true');
+              return;
+            }
+
+            // Skip for scheduled runs — they should collect all failures, not fast-fail
+            if (context.eventName === 'schedule') {
+              core.notice('[health-check] SKIP: scheduled run');
+              return;
+            }
+
+            // Check lint status from the separate Lint workflow (lint.yml).
+            // listJobsForWorkflowRun only sees jobs within the SAME run, so we use
+            // checks.listForRef which queries by commit SHA across ALL workflows.
+            const ref = context.payload.pull_request?.head?.sha || context.sha;
+            core.info(`[health-check] Checking lint for ref=${ref}`);
+            const { data } = await github.rest.checks.listForRef({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              ref: ref,
+              check_name: 'lint',
+            });
+            const lintRun = data.check_runs.find(
+              cr => cr.app?.slug === 'github-actions'
+            );
+            core.info(`[health-check] Lint check: status=${lintRun?.status}, conclusion=${lintRun?.conclusion}`);
+            if (lintRun?.status === 'completed' && lintRun?.conclusion === 'failure') {
+              core.setFailed('Fast-fail: lint check failed');
+              return;
+            }
+
+            // Skip the jobs-failed check when the PR carries the bypass-fastfail label.
+            // Lint check above still runs.
+            let labels = [];
+            if (context.payload.pull_request?.labels) {
+              labels = context.payload.pull_request.labels.map(l => l.name);
+            } else {
+              const { data: prs } = await github.rest.repos.listPullRequestsAssociatedWithCommit({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                commit_sha: ref,
+              });
+              if (prs.length > 0) {
+                labels = prs[0].labels.map(l => l.name);
+              }
+            }
+            core.info(`[health-check] PR labels: [${labels.join(', ')}]`);
+            if (labels.includes('bypass-fastfail')) {
+              core.notice('[health-check] SKIP jobs-failed check: bypass-fastfail label present');
+              return;
+            }
+
+            const jobs = await github.paginate(github.rest.actions.listJobsForWorkflowRun, {
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              run_id: context.runId,
+              per_page: 100,
+            });
+            core.info(`[health-check] Total jobs in run: ${jobs.length}`);
+            const failedJobs = jobs.filter(j => j.status === 'completed' && j.conclusion === 'failure');
+            core.info(`[health-check] Failed jobs (before filtering): ${failedJobs.map(j => `${j.name}(${j.conclusion})`).join(', ') || 'none'}`);
+
+            // Find jobs that failed from a real error, not from fast-fail cascade
+            const rootCauseFailures = jobs.filter(j => {
+              if (j.status !== 'completed' || j.conclusion !== 'failure') return false;
+              // h20 runners are flaky (dirty GPU state from prior runs); their failures
+              // should not cascade fast-fail to other stages. j.name shape from
+              // listJobsForWorkflowRun: "<job-key>" + optional " / <reusable-job>"
+              // + optional " (<matrix>)". Split off the base job key before exact
+              // match so we cover both inline + reusable forms without confusing
+              // 'h20' with the 'h200' prefix.
+              const baseName = j.name.split(/[ /]/)[0];
+              if (baseName === 'base-c-test-8-gpu-h20') {
+                core.info(`[health-check] Filtered out h20 job: ${j.name}`);
+                return false;
+              }
+              // If the failing step is the health check, it's a cascade — skip it
+              const failedStep = (j.steps || []).find(s => s.conclusion === 'failure');
+              if (failedStep && (failedStep.name.includes('check-pr-test-health') || failedStep.name.includes('Check PR test health'))) {
+                core.info(`[health-check] Filtered out cascade failure: ${j.name} (failed step: ${failedStep.name})`);
+                return false;
+              }
+              return true;
+            });
+            core.info(`[health-check] Root cause failures (after filtering): ${rootCauseFailures.map(j => j.name).join(', ') || 'none'}`);
+
+            if (rootCauseFailures.length > 0) {
+              core.setFailed(`Fast-fail: skipping — root cause job(s): ${rootCauseFailures.map(j => j.name).join(', ')}`);
+            } else {
+              core.notice('[health-check] PASS: no root cause failures detected');
+            }
+
+      - name: Install dependencies
+        env:
+          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
+          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
+          UV_INDEX_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
+          RUSTUP_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local:8082"
+        run: |
+          # speed up by using infra cache services
+          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
+          sed -Ei "s@(ports|archive).ubuntu.com@${CACHING_URL}:8081@g" /etc/apt/sources.list
+          pip config set global.index-url http://${CACHING_URL}/pypi/simple
+          pip config set global.trusted-host "${CACHING_URL}"
+
+          bash scripts/ci/npu/npu_ci_install_dependency.sh ${{ inputs.npu_device_type }}
+          # copy required file from our daily cache
+          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
+
+          # install sglang_router
+          apt-get install -y libssl-dev
+          pip install sglang_router
+
+      - name: Run test
+        timeout-minutes: ${{ fromJson(inputs.run_timeout_minutes) }}
+        env:
+          CONTINUE_ON_ERROR_FLAG: ${{ inputs.continue_on_error == 'true' && '--continue-on-error' || '' }}
+        run: |
+          cd test
+          python3 run_suite.py --hw npu --suite ${{ inputs.self_name }} \
+            --auto-partition-id ${{ matrix.partition }} \
+            --auto-partition-size ${{ fromJson(inputs.partitions).size }} \
+            ${{ inputs.timeout_per_file && format('--timeout-per-file {0}', inputs.timeout_per_file) || '' }} \
+            $CONTINUE_ON_ERROR_FLAG
diff --git a/.github/workflows/_npu-single-node-test-stage.yml b/.github/workflows/_npu-single-node-test-stage.yml
new file mode 100644
index 000000000..f04757db7
--- /dev/null
+++ b/.github/workflows/_npu-single-node-test-stage.yml
@@ -0,0 +1,295 @@
+name: 'Single Node Template for E2E performance and accuracy tests'
+
+on:
+  workflow_call:
+    inputs:
+      runner:
+        required: true
+        type: string
+        default: linux-aarch64-a3-16
+      test_type:
+        required: true
+        type: string
+        default: perf
+        description: perf or accuracy
+      test_suite:
+        required: true
+        type: string
+        default: ''
+        description: name of test suite to run via run_suite.py (mutually exclusive with test_case)
+      image:
+        required: true
+        type: string
+        description: image for pods
+        default: "swr.cn-southwest-2.myhuaweicloud.com/base_image/dockerhub/lmsysorg/sglang:main-cann9.0.0-a3"
+      install_sglang_deps:
+        required: false
+        type: boolean
+        default: false
+        description: install sglang dependencies (e.g. PyTorch, CANN packages) when using source installation
+      device_type_for_deps:
+        required: false
+        type: string
+        default: 'a3'
+        description: device type for dependency installation (a3 or 910b)
+      transformers_version:
+        required: false
+        type: string
+        default: ""
+        description: "The transformers version number for running sglang. Use default version in image if keep empty."
+      skip_pr_test_health_check:
+        description: 'Git ref (branch, tag, or SHA) to test. If not provided, uses the default branch.'
+        type: string
+        default: 'false'
+      github-token:
+        description: 'GitHub token for API calls'
+        type: string
+        default: ${{ github.token }}
+
+env:
+  SKIP_PR_TEST_HEALTH_CHECK: ${{ inputs.skip_pr_test_health_check }}
+
+concurrency:
+  group: ascend-nightly-e2e-singlenode-${{ github.workflow_ref }}-${{ github.ref }}-${{ inputs.test_suite }}
+  cancel-in-progress: true
+
+jobs:
+  e2e:
+    name: ${{ inputs.test_suite }}
+    runs-on: ${{ inputs.runner }}
+    container:
+      image: ${{ inputs.image }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Check PR test health
+        uses: actions/github-script@v8
+        env:
+          SKIP_PR_TEST_HEALTH_CHECK: ${{ env.SKIP_PR_TEST_HEALTH_CHECK }}
+        with:
+          github-token: ${{ inputs.github-token || github.token }}
+          script: |
+            core.notice(`[health-check] START — event=${context.eventName}, runId=${context.runId}`);
+
+            // Skip when explicitly requested via env var (e.g. release branch cut)
+            if (process.env.SKIP_PR_TEST_HEALTH_CHECK === 'true') {
+              core.notice('[health-check] SKIP: SKIP_PR_TEST_HEALTH_CHECK=true');
+              return;
+            }
+
+            // Skip for scheduled runs — they should collect all failures, not fast-fail
+            if (context.eventName === 'schedule') {
+              core.notice('[health-check] SKIP: scheduled run');
+              return;
+            }
+
+            // Check lint status from the separate Lint workflow (lint.yml).
+            // listJobsForWorkflowRun only sees jobs within the SAME run, so we use
+            // checks.listForRef which queries by commit SHA across ALL workflows.
+            const ref = context.payload.pull_request?.head?.sha || context.sha;
+            core.info(`[health-check] Checking lint for ref=${ref}`);
+            const { data } = await github.rest.checks.listForRef({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              ref: ref,
+              check_name: 'lint',
+            });
+            const lintRun = data.check_runs.find(
+              cr => cr.app?.slug === 'github-actions'
+            );
+            core.info(`[health-check] Lint check: status=${lintRun?.status}, conclusion=${lintRun?.conclusion}`);
+            if (lintRun?.status === 'completed' && lintRun?.conclusion === 'failure') {
+              core.setFailed('Fast-fail: lint check failed');
+              return;
+            }
+
+            // Skip the jobs-failed check when the PR carries the bypass-fastfail label.
+            // Lint check above still runs.
+            let labels = [];
+            if (context.payload.pull_request?.labels) {
+              labels = context.payload.pull_request.labels.map(l => l.name);
+            } else {
+              const { data: prs } = await github.rest.repos.listPullRequestsAssociatedWithCommit({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                commit_sha: ref,
+              });
+              if (prs.length > 0) {
+                labels = prs[0].labels.map(l => l.name);
+              }
+            }
+            core.info(`[health-check] PR labels: [${labels.join(', ')}]`);
+            if (labels.includes('bypass-fastfail')) {
+              core.notice('[health-check] SKIP jobs-failed check: bypass-fastfail label present');
+              return;
+            }
+
+            const jobs = await github.paginate(github.rest.actions.listJobsForWorkflowRun, {
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              run_id: context.runId,
+              per_page: 100,
+            });
+            core.info(`[health-check] Total jobs in run: ${jobs.length}`);
+            const failedJobs = jobs.filter(j => j.status === 'completed' && j.conclusion === 'failure');
+            core.info(`[health-check] Failed jobs (before filtering): ${failedJobs.map(j => `${j.name}(${j.conclusion})`).join(', ') || 'none'}`);
+
+            // Find jobs that failed from a real error, not from fast-fail cascade
+            const rootCauseFailures = jobs.filter(j => {
+              if (j.status !== 'completed' || j.conclusion !== 'failure') return false;
+              // h20 runners are flaky (dirty GPU state from prior runs); their failures
+              // should not cascade fast-fail to other stages. j.name shape from
+              // listJobsForWorkflowRun: "<job-key>" + optional " / <reusable-job>"
+              // + optional " (<matrix>)". Split off the base job key before exact
+              // match so we cover both inline + reusable forms without confusing
+              // 'h20' with the 'h200' prefix.
+              const baseName = j.name.split(/[ /]/)[0];
+              if (baseName === 'base-c-test-8-gpu-h20') {
+                core.info(`[health-check] Filtered out h20 job: ${j.name}`);
+                return false;
+              }
+              // If the failing step is the health check, it's a cascade — skip it
+              const failedStep = (j.steps || []).find(s => s.conclusion === 'failure');
+              if (failedStep && (failedStep.name.includes('check-pr-test-health') || failedStep.name.includes('Check PR test health'))) {
+                core.info(`[health-check] Filtered out cascade failure: ${j.name} (failed step: ${failedStep.name})`);
+                return false;
+              }
+              return true;
+            });
+            core.info(`[health-check] Root cause failures (after filtering): ${rootCauseFailures.map(j => j.name).join(', ') || 'none'}`);
+
+            if (rootCauseFailures.length > 0) {
+              core.setFailed(`Fast-fail: skipping — root cause job(s): ${rootCauseFailures.map(j => j.name).join(', ')}`);
+            } else {
+              core.notice('[health-check] PASS: no root cause failures detected');
+            }
+
+      - name: Check npu info
+        run: |
+          npu-smi info
+
+      - name: Install sglang dependencies
+        if: ${{ inputs.install_sglang_deps == true }}
+        shell: bash
+        env:
+          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
+          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
+          UV_INDEX_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
+          RUSTUP_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local:8082"
+        run: |
+          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
+          sed -Ei "s@(ports|archive).ubuntu.com@${CACHING_URL}:8081@g" /etc/apt/sources.list
+          pip config set global.index-url http://${CACHING_URL}/pypi/simple
+          pip config set global.trusted-host "${CACHING_URL}"
+          bash scripts/ci/npu/npu_ci_install_dependency.sh ${{ inputs.device_type_for_deps }}
+          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
+          curl -o /tmp/test.jsonl -L https://gh-proxy.test.osinfra.cn/https://raw.githubusercontent.com/openai/grade-school-math/master/grade_school_math/data/test.jsonl
+
+      - name: Run test
+        timeout-minutes: 300
+        env:
+          SGLANG_USE_MODELSCOPE: true
+          HF_ENDPOINT: https://hf-mirror.com
+          SGLANG_IS_IN_CI: true
+          TRANSFORMERS_VERBOSITY: "error"
+          GDN_ATTN_BACKEND_TRITON: 1
+          SGLANG_TEST_METRICS_OUTPUT: /root/.cache/tests/output/metrics/metrics
+        shell: bash
+        run: |
+          sglang_source_path=$(pwd)
+          echo "Source code path: ${sglang_source_path}"
+          ln -sf ${sglang_source_path} /root/sglang
+
+          # Determine test mode: suite (run_suite.py --suite) or single case file.
+          test_suite="${{ inputs.test_suite }}"
+          echo "Test mode: suite (${test_suite})"
+          tc_name="${test_suite}"
+
+          export TRANSFORMERS_VERSION_FOR_SGLANG="${{ inputs.transformers_version }}"
+          PYTHON_FOR_SGLANG="python"
+          PIP_FOR_SGLANG="pip"
+          if [ -n "${TRANSFORMERS_VERSION_FOR_SGLANG}" ];then
+            echo "===== Install transformers for sglang - Begin ====="
+            TRANSFORMERS_PKG_PATH_SOURCE=/root/.cache/.cache/transformers/${TRANSFORMERS_VERSION_FOR_SGLANG}
+            if [ ! -d "${TRANSFORMERS_PKG_PATH_SOURCE}" ]; then
+              echo "The dependent transformers package does not exist: ${TRANSFORMERS_PKG_PATH_SOURCE}."
+              echo "Install transformers ${TRANSFORMERS_VERSION_FOR_SGLANG} online."
+              pip install transformers=="${TRANSFORMERS_VERSION_FOR_SGLANG}" -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
+            else
+              echo "Install transformers ${TRANSFORMERS_VERSION_FOR_SGLANG} locally."
+              TRANSFORMERS_PKG_PATH_TARGET=/tmp/transformers/${TRANSFORMERS_VERSION_FOR_SGLANG}
+              mkdir -p "${TRANSFORMERS_PKG_PATH_TARGET}"
+              cp "${TRANSFORMERS_PKG_PATH_SOURCE}/*" "${TRANSFORMERS_PKG_PATH_TARGET}/"
+              pip install --no-index --find-links="${TRANSFORMERS_PKG_PATH_TARGET}" transformers=="${TRANSFORMERS_VERSION_FOR_SGLANG}"
+            fi
+            echo "===== Install transformers for sglang in virtual env - End ====="
+          fi
+          echo "Transformers version for sglang: $(${PIP_FOR_SGLANG} show transformers | grep Version | cut -d: -f2)"
+
+          echo "scaling_governor performance num: \
+            $(cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor | grep performance | wc -l)"
+          echo "swappiness: $(cat /proc/sys/vm/swappiness)"
+          echo "numa_balancing: $(cat /proc/sys/kernel/numa_balancing)"
+          echo "sched_migration_cost_ns: $(cat /proc/sys/kernel/sched_migration_cost_ns)"
+
+          export SGLANG_TEST_MAX_RETRY=0
+          export SGLANG_SET_CPU_AFFINITY=1
+          echo "SGLANG_SET_CPU_AFFINITY: $SGLANG_SET_CPU_AFFINITY"
+
+
+          # Copy the checked‑out test common utility code to the sglang installation directory.
+          sglang_pkg_path=/sgl-workspace/sglang/python
+          ascend_test_util_path=${sglang_pkg_path}/sglang/test/ascend
+          mkdir -p ${ascend_test_util_path}
+          mv ${ascend_test_util_path} ${ascend_test_util_path}_bak
+          cp -r ${sglang_source_path}/python/sglang/test/ascend ${ascend_test_util_path}
+
+          source /usr/local/Ascend/cann/set_env.sh || true
+          source /usr/local/Ascend/nnal/atb/set_env.sh || true
+          source /usr/local/Ascend/ascend-toolkit/latest/opp/vendors/customize/bin/set_env.bash || true
+          source /usr/local/Ascend/ascend-toolkit/latest/opp/vendors/custom_transformer/bin/set_env.bash || true
+
+          # Set environment of cann
+          log_path="/root/.cache/tests/logs/log/${current_date}/${tc_name}/${HOSTNAME}"
+          rm -rf ${log_path}
+          mkdir -p ${log_path}
+          echo "Log path: ${log_path}"
+
+          echo "Running test: ${tc_name}"
+          test_exit_code=0
+          cd test
+          ${PYTHON_FOR_SGLANG} -u run_suite.py --hw npu --suite ${test_suite} --timeout-per-file 3600 2>&1 | tee /tmp/test_output.log || test_exit_code=$?
+
+          echo "Finished test: ${tc_name}"
+
+          if [ "${test_exit_code}" = "0" ]; then
+            test_status="pass"
+            status_icon="✅"
+          else
+            test_status="fail"
+            status_icon="❌"
+          fi
+
+          echo "test_status=${test_status}" >> $GITHUB_ENV
+          echo "tc_name=${tc_name}" >> $GITHUB_ENV
+          export test_status tc_name
+
+          echo "## ${tc_name}  ${status_icon} ${test_status}" >> $GITHUB_STEP_SUMMARY
+          echo "" >> $GITHUB_STEP_SUMMARY
+          metric_count=$(grep -c '\[METRIC\]' /tmp/test_output.log 2>/dev/null || echo 0)
+          if [ "${metric_count}" -gt 0 ]; then
+            echo "| Metric | Value | Pass |" >> $GITHUB_STEP_SUMMARY
+            echo "|--------|-------|------|" >> $GITHUB_STEP_SUMMARY
+            grep '\[METRIC\]' /tmp/test_output.log | while IFS= read -r line; do
+              metric_name=$(echo "$line" | sed -E 's/.*\[METRIC\] ([^=]+)=.*/\1/')
+              metric_value=$(echo "$line" | sed -E 's/.*\[METRIC\] [^=]+=([^ ]+).*/\1/')
+              echo "| ${metric_name} | ${metric_value} | ${status_icon} |" >> $GITHUB_STEP_SUMMARY
+            done
+          else
+            echo "No metrics collected (test may have failed before producing results)." >> $GITHUB_STEP_SUMMARY
+          fi
+          echo "" >> $GITHUB_STEP_SUMMARY
+
+          exit ${test_exit_code}
diff --git a/.github/workflows/pr-test-npu.yml b/.github/workflows/pr-test-npu.yml
index 3387c6b78..f829631f4 100644
--- a/.github/workflows/pr-test-npu.yml
+++ b/.github/workflows/pr-test-npu.yml
@@ -90,303 +90,82 @@ jobs:
           echo "CANN_image_a3=swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:9.0.0-a3-ubuntu22.04-py3.11" >> $GITHUB_OUTPUT
           echo "CANN_image_910b=swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:9.0.0-910b-ubuntu22.04-py3.11" >> $GITHUB_OUTPUT
 
-  stage-a-unit-test-npu:
+  base-a-test-1-npu-a2:
     needs: [check-changes, pr-gate, set-image-config]
-    if: needs.check-changes.outputs.main_package == 'true'
-    runs-on: linux-aarch64-a2-1
-    container:
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-pr-test-stage.yml
+    with:
+      self_name:  base-a-test-1-npu-a2
+      runner_config: linux-aarch64-a2-1
       image: ${{ needs.set-image-config.outputs.CANN_image_910b }}
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
+      npu_device_type: 910b
+      run_timeout_minutes: '15'
+    secrets: inherit
 
-      - name: Mark repository safe
-        run: |
-          git config --system --add safe.directory ${GITHUB_WORKSPACE}
-
-      - name: Install dependencies
-        env:
-          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
-          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          UV_INDEX_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
-          RUSTUP_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local:8082"
-        run: |
-          # speed up by using infra cache services
-          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
-          sed -Ei "s@(ports|archive).ubuntu.com@${CACHING_URL}:8081@g" /etc/apt/sources.list
-          pip config set global.index-url http://${CACHING_URL}/pypi/simple
-          pip config set global.trusted-host "${CACHING_URL}"
-
-          bash scripts/ci/npu/npu_ci_install_dependency.sh 910b
-
-      - name: Run test
-        timeout-minutes: 15
-        env:
-          SGLANG_IS_IN_CI: true
-        run: |
-          cd test
-          python3 run_suite.py --hw npu --suite stage-a-unit-test-npu
-
-  stage-b-test-1-npu-a3:
+  base-b-test-1-npu-a3:
     needs: [check-changes, pr-gate, set-image-config]
-    if: needs.check-changes.outputs.main_package == 'true'
-    runs-on: linux-aarch64-a3-2-
-    container:
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-pr-test-stage.yml
+    with:
+      self_name: base-b-test-1-npu-a3
+      runner_config: linux-aarch64-a3-2-
       image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
+      run_timeout_minutes: '60'
+      timeout_per_file: '3600'
+    secrets: inherit
 
-      - name: Mark repository safe
-        run: |
-          git config --system --add safe.directory ${GITHUB_WORKSPACE}
-
-      - name: Install dependencies
-        env:
-          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
-          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          UV_INDEX_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
-          RUSTUP_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local:8082"
-        run: |
-          # speed up by using infra cache services
-          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
-          sed -Ei "s@(ports|archive).ubuntu.com@${CACHING_URL}:8081@g" /etc/apt/sources.list
-          pip config set global.index-url http://${CACHING_URL}/pypi/simple
-          pip config set global.trusted-host "${CACHING_URL}"
-
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
-          # copy required file from our daily cache
-          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy gsm8k dataset
-          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
-
-      - name: Run test
-        timeout-minutes: 60
-        env:
-          SGLANG_USE_MODELSCOPE: true
-          SGLANG_IS_IN_CI: true
-          HF_ENDPOINT: https://hf-mirror.com
-          TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
-          PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
-          STREAMS_PER_DEVICE: 32
-        run: |
-          cd test
-          python3 run_suite.py --hw npu --suite stage-b-test-1-npu-a3 --timeout-per-file 3600
-
-
-  stage-b-test-2-npu-a3:
+  base-b-test-2-npu-a3:
     needs: [check-changes, pr-gate, set-image-config]
-    if: needs.check-changes.outputs.main_package == 'true'
-    runs-on: linux-aarch64-a3-2-
-    container:
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-pr-test-stage.yml
+    with:
+      self_name: base-b-test-2-npu-a3
+      runner_config: linux-aarch64-a3-2-
       image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
+      run_timeout_minutes: '60'
+      timeout_per_file: '3600'
+    secrets: inherit
 
-      - name: Mark repository safe
-        run: |
-          git config --system --add safe.directory ${GITHUB_WORKSPACE}
-
-      - name: Install dependencies
-        env:
-          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
-          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          UV_INDEX_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
-          RUSTUP_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local:8082"
-        run: |
-          # speed up by using infra cache services
-          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
-          sed -Ei "s@(ports|archive).ubuntu.com@${CACHING_URL}:8081@g" /etc/apt/sources.list
-          pip config set global.index-url http://${CACHING_URL}/pypi/simple
-          pip config set global.trusted-host "${CACHING_URL}"
-
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
-          # copy required file from our daily cache
-          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy gsm8k dataset
-          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
-
-      - name: Run test
-        timeout-minutes: 60
-        env:
-          SGLANG_USE_MODELSCOPE: true
-          SGLANG_IS_IN_CI: true
-          HF_ENDPOINT: https://hf-mirror.com
-          TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
-          PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
-          STREAMS_PER_DEVICE: 32
-        run: |
-          cd test
-          python3 run_suite.py --hw npu --suite stage-b-test-2-npu-a3 --timeout-per-file 3600
-
-
-  stage-b-test-4-npu-a3:
+  base-b-test-4-npu-a3:
     needs: [check-changes, pr-gate, set-image-config]
-    if: needs.check-changes.outputs.main_package == 'true'
-    runs-on: linux-aarch64-a3-4-
-    strategy:
-      fail-fast: false
-      matrix:
-        part: [ 0, 1 ]
-    container:
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-pr-test-stage.yml
+    with:
+      self_name: base-b-test-4-npu-a3
+      runner_config: linux-aarch64-a3-4-
       image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
+      run_timeout_minutes: '120'
+      timeout_per_file: '3600'
+      partitions: '{"size":2,"arr":[0, 1]}'
+    secrets: inherit
 
-      - name: Mark repository safe
-        run: |
-          git config --system --add safe.directory ${GITHUB_WORKSPACE}
-
-      - name: Install dependencies
-        env:
-          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
-          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          UV_INDEX_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
-          RUSTUP_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local:8082"
-        run: |
-          # speed up by using infra cache services
-          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
-          sed -Ei "s@(ports|archive).ubuntu.com@${CACHING_URL}:8081@g" /etc/apt/sources.list
-          pip config set global.index-url http://${CACHING_URL}/pypi/simple
-          pip config set global.trusted-host "${CACHING_URL}"
-
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
-          # copy required file from our daily cache
-          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy gsm8k dataset
-          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
-
-      - name: Run test
-        timeout-minutes: 120
-        env:
-          SGLANG_USE_MODELSCOPE: true
-          SGLANG_IS_IN_CI: true
-          HF_ENDPOINT: https://hf-mirror.com
-          TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
-          PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
-          STREAMS_PER_DEVICE: 32
-        run: |
-          pip install sglang_router
-          cd test
-          python3 run_suite.py --hw npu --suite stage-b-test-4-npu-a3 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2 --timeout-per-file 3600
-
-  stage-b-test-8-npu-a3:
+  base-b-test-8-npu-a3:
     needs: [check-changes, pr-gate, set-image-config]
-    if: needs.check-changes.outputs.main_package == 'true'
-    runs-on: linux-aarch64-a3-8-
-    container:
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-pr-test-stage.yml
+    with:
+      self_name: base-b-test-8-npu-a3
+      runner_config: linux-aarch64-a3-8-
       image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
+      run_timeout_minutes: '60'
+      timeout_per_file: '3600'
+    secrets: inherit
 
-      - name: Mark repository safe
-        run: |
-          git config --system --add safe.directory ${GITHUB_WORKSPACE}
-
-      - name: Install dependencies
-        env:
-          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
-          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          UV_INDEX_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
-          RUSTUP_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local:8082"
-        run: |
-          # speed up by using infra cache services
-          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
-          sed -Ei "s@(ports|archive).ubuntu.com@${CACHING_URL}:8081@g" /etc/apt/sources.list
-          pip config set global.index-url http://${CACHING_URL}/pypi/simple
-          pip config set global.trusted-host "${CACHING_URL}"
-
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
-          # copy required file from our daily cache
-          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy gsm8k dataset
-          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
-
-      - name: Run test
-        timeout-minutes: 60
-        env:
-          SGLANG_USE_MODELSCOPE: true
-          SGLANG_IS_IN_CI: true
-          HF_ENDPOINT: https://hf-mirror.com
-          TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
-          PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
-          STREAMS_PER_DEVICE: 32
-        run: |
-          cd test
-          python3 run_suite.py --hw npu --suite stage-b-test-8-npu-a3 --timeout-per-file 3600
-
-  stage-b-test-16-npu-a3:
+  base-b-test-16-npu-a3:
     needs: [check-changes, pr-gate, set-image-config]
-    if: needs.check-changes.outputs.main_package == 'true'
-    runs-on: linux-aarch64-a3-16-
-    container:
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-pr-test-stage.yml
+    with:
+      self_name: base-b-test-16-npu-a3
+      runner_config: linux-aarch64-a3-16-
       image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.ref || github.ref }}
-
-      - name: Mark repository safe
-        run: |
-          git config --system --add safe.directory ${GITHUB_WORKSPACE}
-
-      - name: Install dependencies
-        env:
-          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
-          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          UV_INDEX_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
-          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
-          RUSTUP_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local:8082"
-        run: |
-          # speed up by using infra cache services
-          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
-          sed -Ei "s@(ports|archive).ubuntu.com@${CACHING_URL}:8081@g" /etc/apt/sources.list
-          pip config set global.index-url http://${CACHING_URL}/pypi/simple
-          pip config set global.trusted-host "${CACHING_URL}"
-
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
-          # copy required file from our daily cache
-          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy gsm8k dataset
-          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
-
-      - name: Run test
-        timeout-minutes: 120
-        env:
-          SGLANG_USE_MODELSCOPE: true
-          SGLANG_IS_IN_CI: true
-          HF_ENDPOINT: https://hf-mirror.com
-          TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
-          PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
-          STREAMS_PER_DEVICE: 32
-        run: |
-          apt-get install -y libssl-dev
-          pip install sglang_router
-          cd test
-          python3 run_suite.py --hw npu --suite stage-b-test-16-npu-a3 --timeout-per-file 3600
+      run_timeout_minutes: '120'
+      timeout_per_file: '3600'
+    secrets: inherit
 
   multimodal-gen-test-1-npu-a3:
     needs: [check-changes, pr-gate, set-image-config]
-    if: needs.check-changes.outputs.multimodal_gen == 'true'
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.multimodal_gen == 'true' }}
     runs-on: linux-aarch64-a3-800t-2
     container:
       image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
@@ -443,7 +222,7 @@ jobs:
 
   multimodal-gen-test-2-npu-a3:
     needs: [check-changes, pr-gate, set-image-config]
-    if: needs.check-changes.outputs.multimodal_gen == 'true'
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.multimodal_gen == 'true' }}
     runs-on: linux-aarch64-a3-800t-2
     container:
       image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
@@ -498,146 +277,134 @@ jobs:
           if-no-files-found: ignore
           retention-days: 7
 
-  pr-single-node-tests:
-    name: single-node-poc
-    needs: [check-changes, pr-gate, set-image-config]
-    if: needs.check-changes.outputs.main_package == 'true'
-    strategy:
-      fail-fast: false
-      max-parallel: 6
-      matrix:
-        test_config:
-          # qwen3_6_27b performance tests
-          - name: qwen3_6_27b_w8a8_1p_in64k_out1k_50ms
-            runner: linux-aarch64-a3-800t-2
-            test_case: test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_w8a8_1p_in64k_out1k_50ms.py
-            test_type: 'perf'
-
-          # - name: qwen3_6_27b_1p_in1024x1024_30_out1024_50ms
-          #   runner: linux-aarch64-a3-800t-2
-          #   test_case: test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_1p_in1024x1024_30_out1024_50ms.py
-          #   test_type: 'perf'
-
-          # # qwen3_8b performance tests
-          # - name: qwen3_8b_w8a8_1p_in3k5_out1k5_50ms
-          #   runner: linux-aarch64-a3-800t-2
-          #   test_case: test/registered/npu/performance/qwen3-8b/test_npu_qwen3_8b_w8a8_1p_in3k5_out1k5_50ms.py
-          #   test_type: 'perf'
-
-          # # qwen3_30b_a3b performance tests
-          # - name: qwen3_30b_w8a8_1p_in3k5_out1k5_50ms
-          #   runner: linux-aarch64-a3-800t-2
-          #   test_case: test/registered/npu/performance/qwen3_30b_a3b/test_npu_qwen3_30b_w8a8_1p_in3k5_out1k5_50ms.py
-          #   test_type: 'perf'
-
-          # # qwen3_6_35b_a3b performance tests
-          # - name: qwen3_6_35b_a3b_1p_in64k_out1k_prefix90_50ms
-          #   runner: linux-aarch64-a3-800t-2
-          #   test_case: test/registered/npu/performance/qwen3_6_35b_a3b/test_npu_qwen3_6_35b_a3b_1p_in64k_out1k_prefix90_50ms.py
-          #   test_type: 'perf'
-
-          # # qwen3_vl_8b_thinking accuracy tests
-          # - name: qwen3_vl_8b_thinking_1p_mmmu
-          #   runner: linux-aarch64-a3-2-
-          #   test_case: test/registered/npu/accuracy/qwen3_vl_8b_thinking/test_npu_qwen3_vl_8b_thinking_1p_mmmu.py
-          #   test_type: 'accuracy'
-
-          # # qwen3_32b performance tests
-          # - name: qwen3_32b_w8a8_2p_in3k5_out1k5_50ms
-          #   runner: linux-aarch64-a3-800t-4
-          #   test_case: test/registered/npu/performance/qwen3_32b/test_npu_qwen3_32b_w8a8_2p_in3k5_out1k5_50ms.py
-          #   test_type: 'perf'
-
-          # # qwen3_next_80b_a3b performance tests
-          # - name: qwen3_next_80b_w8a8_2p_in6k_out1k5_bs16
-          #   runner: linux-aarch64-a3-800t-4
-          #   test_case: test/registered/npu/performance/qwen3_next_80b_a3b_instruct/test_npu_qwen3_next_80b_w8a8_2p_in6k_out1k5_bs16.py
-          #   test_type: 'perf'
-
-          # # minimax_m2_5 performance tests
-          # - name: minimax_m2_5_w8a8_4p_in64k_out1k_prefix90_50ms
-          #   runner: linux-aarch64-a3-800t-8
-          #   test_case: test/registered/npu/performance/minimax_m2_5/test_npu_minimax_m2_5_w8a8_4p_in64k_out1k_prefix90_50ms.py
-          #   test_type: 'perf'
-
-          # # deepseek_v4_flash performance tests
-          # - name: deepseek_v4_flash_w8a8_8p_in8k_out1k_50ms
-          #   runner: linux-aarch64-a3-800t-16
-          #   test_case: test/registered/npu/performance/deepseek_v4_flash/test_npu_deepseek_v4_flash_w8a8_8p_in8k_out1k_50ms.py
-          #   test_type: 'perf'
-
-          # # kimi_k2_6 performance tests
-          # - name: kimi_k2_6_w4a8_8p_in3k5_out1k5_20ms
-          #   runner: linux-aarch64-a3-800t-16
-          #   test_case: test/registered/npu/performance/kimi_k2_6/test_npu_kimi_k2_6_w4a8_8p_in3k5_out1k5_20ms.py
-          #   test_type: 'perf'
-
-          # # qwen3_235b performance tests
-          # - name: qwen3_235b_w8a8_8p_in3k5_out1k5_50ms
-          #   runner: linux-aarch64-a3-800t-16
-          #   test_case: test/registered/npu/performance/qwen3_235b_a22b/test_npu_qwen3_235b_w8a8_8p_in3k5_out1k5_50ms.py
-          #   test_type: 'perf'
-
-          # # qwen3_5_397b performance tests
-          # - name: qwen3_5_397b_w4a8_8p_in3k5_out1k5_50ms
-          #   runner: linux-aarch64-a3-800t-16
-          #   test_case: test/registered/npu/performance/qwen3_5_397b/test_npu_qwen3_5_397b_w4a8_8p_in3k5_out1k5_50ms.py
-          #   test_type: 'perf'
-
-          # NPU accuracy tests
-          # - name: glm4_7_flash_1p_gsm8k
-          #   runner: linux-aarch64-a3-2-
-          #   test_case: test/registered/npu/accuracy/glm4_7_flash/test_npu_glm4_7_flash_1p_gsm8k.py
-          #   test_type: 'accuracy'
-          # - name: qwen3_vl_30b_a3b_bf16_2p_gsm8k
-          #   runner: linux-aarch64-a3-4-
-          #   test_case: test/registered/npu/accuracy/qwen3_vl_30b_a3b/test_npu_qwen3_vl_30b_a3b_bf16_2p_gsm8k.py
-          #   test_type: 'accuracy'
-          - name: glm5_top64_pruned_bf16_8p_gsm8k
-            runner: linux-aarch64-a3-16-
-            test_case: test/registered/npu/accuracy/glm5_top64_pruned/test_npu_glm5_top64_pruned_bf16_8p_gsm8k.py
-            test_type: 'accuracy'
-          # - name: moonshotai_moonlight_16b_a3b_bf16_1p_gsm8k
-          #   runner: linux-aarch64-a3-2-
-          #   test_case: test/registered/npu/accuracy/moonshotai_moonlight_16b_a3b/test_npu_moonlight_16b_a3b_bf16_1p_gsm8k.py
-          #   test_type: 'accuracy'
-          # - name: qwen3_5_9b_bf16_1p_gsm8k
-          #   runner: linux-aarch64-a3-2-
-          #   test_case: test/registered/npu/accuracy/qwen3_5_9b/test_npu_qwen3_5_9b_bf16_1p_gsm8k.py
-          #   test_type: 'accuracy'
-          # - name: qwen3_vl_8b_bf16_2p_gsm8k
-          #   runner: linux-aarch64-a3-4-
-          #   test_case: test/registered/npu/accuracy/qwen3_vl_8b/test_npu_qwen3_vl_8b_bf16_2p_gsm8k.py
-          #   test_type: 'accuracy'
-
-    uses: ./.github/workflows/nightly-test-npu-e2e-single-node.yml
+  base-c-test-acc-2-npu-a3:
+    name: base-c-test-acc-2-npu-a3
+    needs: [ check-changes, pr-gate, set-image-config ]
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-single-node-test-stage.yml
     with:
-      runner: ${{ matrix.test_config.runner }}
-      test_type: ${{ matrix.test_config.test_type }}
-      test_config_name: ${{ matrix.test_config.name }}
-      test_case: ${{ matrix.test_config.test_case }}
+      runner: linux-aarch64-a3-2-
+      test_type: 'accuracy'
+      test_suite: base-c-test-acc-2-npu-a3
       image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
-      install_sglang_from_source: false
       install_sglang_deps: true
       device_type_for_deps: 'a3'
-      transformers_version: ''
+
+  base-c-test-acc-4-npu-a3:
+    name: base-c-test-acc-4-npu-a3
+    needs: [ check-changes, pr-gate, set-image-config ]
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-single-node-test-stage.yml
+    with:
+      runner: linux-aarch64-a3-4-
+      test_type: 'accuracy'
+      test_suite: base-c-test-acc-4-npu-a3
+      image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
+      install_sglang_deps: true
+      device_type_for_deps: 'a3'
+
+  base-c-test-acc-8-npu-a3:
+    name: base-c-test-acc-8-npu-a3
+    needs: [ check-changes, pr-gate, set-image-config ]
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-single-node-test-stage.yml
+    with:
+      runner: linux-aarch64-a3-8-
+      test_type: 'accuracy'
+      test_suite: base-c-test-acc-8-npu-a3
+      image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
+      install_sglang_deps: true
+      device_type_for_deps: 'a3'
+
+  base-c-test-acc-16-npu-a3:
+    name: base-c-test-acc-16-npu-a3
+    needs: [ check-changes, pr-gate, set-image-config ]
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-single-node-test-stage.yml
+    with:
+      runner: linux-aarch64-a3-16-
+      test_type: 'accuracy'
+      test_suite: base-c-test-acc-16-npu-a3
+      image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
+      install_sglang_deps: true
+      device_type_for_deps: 'a3'
+
+  base-c-test-perf-2-npu-a3:
+    name: base-c-test-perf-2-npu-a3
+    needs: [ check-changes, pr-gate, set-image-config, base-c-test-acc-2-npu-a3 ]
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-single-node-test-stage.yml
+    with:
+      runner: linux-aarch64-a3-800t-2
+      test_type: 'perf'
+      test_suite: base-c-test-perf-2-npu-a3
+      image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
+      install_sglang_deps: true
+      device_type_for_deps: 'a3'
+
+  base-c-test-perf-4-npu-a3:
+    name: base-c-test-perf-4-npu-a3
+    needs: [ check-changes, pr-gate, set-image-config, base-c-test-acc-4-npu-a3 ]
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-single-node-test-stage.yml
+    with:
+      runner: linux-aarch64-a3-800t-4
+      test_type: 'perf'
+      test_suite: base-c-test-perf-4-npu-a3
+      image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
+      install_sglang_deps: true
+      device_type_for_deps: 'a3'
+
+  base-c-test-perf-8-npu-a3:
+    name: base-c-test-perf-8-npu-a3
+    needs: [ check-changes, pr-gate, set-image-config, base-c-test-acc-8-npu-a3 ]
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-single-node-test-stage.yml
+    with:
+      runner: linux-aarch64-a3-800t-8
+      test_type: 'perf'
+      test_suite: base-c-test-perf-8-npu-a3
+      image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
+      install_sglang_deps: true
+      device_type_for_deps: 'a3'
+
+  base-c-test-perf-16-npu-a3:
+    name: base-c-test-perf-16-npu-a3
+    needs: [ check-changes, pr-gate, set-image-config, base-c-test-acc-16-npu-a3 ]
+    if: ${{ !failure() && !cancelled() && needs.check-changes.outputs.main_package == 'true' }}
+    uses: ./.github/workflows/_npu-single-node-test-stage.yml
+    with:
+      runner: linux-aarch64-a3-800t-16
+      test_type: 'perf'
+      test_suite: base-c-test-perf-16-npu-a3
+      image: ${{ needs.set-image-config.outputs.CANN_image_a3 }}
+      install_sglang_deps: true
+      device_type_for_deps: 'a3'
+
 
   pr-test-npu-finish:
     needs:
       [
         check-changes,
 
-        stage-a-unit-test-npu,
-        stage-b-test-1-npu-a3,
-        stage-b-test-2-npu-a3,
-        stage-b-test-4-npu-a3,
-        stage-b-test-8-npu-a3,
-        stage-b-test-16-npu-a3,
+        base-a-test-1-npu-a2,
+        base-b-test-1-npu-a3,
+        base-b-test-2-npu-a3,
+        base-b-test-4-npu-a3,
+        base-b-test-8-npu-a3,
+        base-b-test-16-npu-a3,
 
         multimodal-gen-test-1-npu-a3,
         multimodal-gen-test-2-npu-a3,
 
-        pr-single-node-tests,
+        base-c-test-acc-2-npu-a3,
+        base-c-test-acc-4-npu-a3,
+        base-c-test-acc-8-npu-a3,
+        base-c-test-acc-16-npu-a3,
+        base-c-test-perf-2-npu-a3,
+        base-c-test-perf-4-npu-a3,
+        base-c-test-perf-8-npu-a3,
+        base-c-test-perf-16-npu-a3,
       ]
     if: always()
     runs-on: ubuntu-latest
diff --git a/python/sglang/test/ascend/e2e/test_npu_accuracy_utils.py b/python/sglang/test/ascend/e2e/test_npu_accuracy_utils.py
index 6dc75173a..5fcdf3d5c 100644
--- a/python/sglang/test/ascend/e2e/test_npu_accuracy_utils.py
+++ b/python/sglang/test/ascend/e2e/test_npu_accuracy_utils.py
@@ -1,10 +1,15 @@
+import glob
+import inspect
 import json
 import logging
 import os
 import re
+import shutil
 import subprocess
+import sys
 import threading
 import time
+from datetime import datetime
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
@@ -289,9 +294,117 @@ class TestNpuAccuracyTestCaseBase(CustomTestCase):
     max_attempts = 2
     n_runs = 3
     accuracy = 0.1
+    test_type = "accuracy"
+
+    @classmethod
+    def _get_tc_name(cls):
+        """Derive the test case name from the test file (filename without
+        extension). Mirrors the workflow's ``tc_name=${test_case##*/}`` logic
+        so each case in a suite writes to its own output path."""
+        try:
+            tc_file = inspect.getfile(cls)
+        except (TypeError, OSError):
+            tc_file = getattr(sys.modules.get(cls.__module__), "__file__", "")
+        return os.path.splitext(os.path.basename(tc_file))[0]
+
+    @classmethod
+    def _setup_per_case_output(cls):
+        """Set up per-case output directories and env vars.
+
+        Extracted from ``nightly-test-npu-e2e-single-node.yml`` so that when a
+        suite is executed, each case writes its metrics/plog to a path derived
+        from the case file rather than the suite name.
+        """
+        cls.tc_name = cls._get_tc_name()
+        current_date = datetime.now().strftime("%Y%m%d")
+        test_type = getattr(cls, "test_type", "accuracy")
+        base_output = f"/root/.cache/tests/output/{test_type}/{current_date}"
+        os.makedirs(base_output, exist_ok=True)
+        cls.metrics_data_file = os.path.join(base_output, cls.tc_name)
+        os.makedirs(cls.metrics_data_file, exist_ok=True)
+        # Override env vars so evalscope/dump_metric write to per-case paths.
+        os.environ["METRICS_DATA_FILE"] = cls.metrics_data_file
+        os.environ["SGLANG_TEST_METRICS_OUTPUT"] = os.path.join(
+            cls.metrics_data_file, "metrics"
+        )
+        logger.info(
+            "Per-case output: tc_name=%s metrics_data_file=%s",
+            cls.tc_name,
+            cls.metrics_data_file,
+        )
+
+    @classmethod
+    def _save_metrics_json(cls):
+        """Write per-case ``metrics.json`` from ``dump_metric`` JSONL files.
+
+        Replaces the workflow's stdout-parsing + ``dump_metrics.py`` logic so
+        each case in a suite persists its own metrics snapshot.
+        """
+        if not getattr(cls, "metrics_data_file", None):
+            return
+        metrics = {}
+        baselines = {}
+        pattern = os.path.join(cls.metrics_data_file, "metrics.*.jsonl")
+        for jsonl_path in glob.glob(pattern):
+            try:
+                with open(jsonl_path, "r", encoding="utf-8") as f:
+                    for line in f:
+                        line = line.strip()
+                        if not line:
+                            continue
+                        record = json.loads(line)
+                        name = record.get("metric_name")
+                        value = record.get("value")
+                        if name is None:
+                            continue
+                        if name.endswith("_baseline"):
+                            baselines[name[: -len("_baseline")]] = value
+                        else:
+                            metrics[name] = value
+            except Exception as e:
+                logger.warning("Failed to read %s: %s", jsonl_path, e)
+        out_path = os.path.join(cls.metrics_data_file, "metrics.json")
+        payload = {
+            "test_case": cls.tc_name,
+            "test_type": getattr(cls, "test_type", "accuracy"),
+            "metrics": metrics,
+            "baselines": baselines,
+        }
+        try:
+            with open(out_path, "w", encoding="utf-8") as f:
+                json.dump(payload, f, ensure_ascii=False)
+            logger.info("Saved per-case metrics to %s", out_path)
+        except Exception as e:
+            logger.warning("Failed to write metrics.json: %s", e)
+
+    @classmethod
+    def _backup_plog(cls):
+        """Backup Ascend plog files to a per-case path.
+
+        Replaces the workflow's ``Backup plog`` step so each case in a suite
+        gets its own plog snapshot instead of all cases sharing the suite name.
+        """
+        plog_path = "/root/ascend/log/debug/plog"
+        if not os.path.isdir(plog_path):
+            return
+        tc_name = getattr(cls, "tc_name", None)
+        if not tc_name:
+            return
+        hostname = os.getenv("HOSTNAME", "unknown")
+        target = os.path.join("/root/.cache/tests/logs/plog", tc_name, hostname)
+        os.makedirs(target, exist_ok=True)
+        for name in os.listdir(plog_path):
+            src = os.path.join(plog_path, name)
+            if os.path.isfile(src):
+                try:
+                    shutil.copy2(src, os.path.join(target, name))
+                except Exception as e:
+                    logger.warning("Failed to copy plog %s: %s", name, e)
+        logger.info("Backed up plog to %s", target)
 
     @classmethod
     def setUpClass(cls):
+        cls._setup_per_case_output()
         cls.base_url = DEFAULT_URL_FOR_TEST
         env = os.environ.copy()
         for key, value in env.items():
@@ -318,6 +431,8 @@ class TestNpuAccuracyTestCaseBase(CustomTestCase):
                 kill_process_tree(cls.process.pid)
             except Exception as e:
                 logger.error(f"Error during tearDown: {e}")
+        cls._save_metrics_json()
+        cls._backup_plog()
 
     def run_accuracy(self):
         parsed_url = urlparse(self.base_url)
diff --git a/python/sglang/test/ascend/e2e/test_npu_performance_utils.py b/python/sglang/test/ascend/e2e/test_npu_performance_utils.py
index 21cd57171..dfa30e6b8 100644
--- a/python/sglang/test/ascend/e2e/test_npu_performance_utils.py
+++ b/python/sglang/test/ascend/e2e/test_npu_performance_utils.py
@@ -1,9 +1,15 @@
+import glob
+import inspect
+import json
 import logging
 import os
 import re
+import shutil
 import subprocess
+import sys
 import threading
 import time
+from datetime import datetime
 from functools import wraps
 from urllib.parse import urlparse
 
@@ -909,8 +915,115 @@ class TestNpuPerformanceTestCaseBase(CustomTestCase):
     generation_kwargs = None
     pop_sglang_is_in_ci_for_gsp = False
 
+    @classmethod
+    def _get_tc_name(cls):
+        """Derive the test case name from the test file (filename without
+        extension). Mirrors the workflow's ``tc_name=${test_case##*/}`` logic
+        so each case in a suite writes to its own output path."""
+        try:
+            tc_file = inspect.getfile(cls)
+        except (TypeError, OSError):
+            tc_file = getattr(sys.modules.get(cls.__module__), "__file__", "")
+        return os.path.splitext(os.path.basename(tc_file))[0]
+
+    @classmethod
+    def _setup_per_case_output(cls):
+        """Set up per-case output directories and env vars.
+
+        Extracted from ``nightly-test-npu-e2e-single-node.yml`` so that when a
+        suite is executed, each case writes its metrics/plog to a path derived
+        from the case file rather than the suite name.
+        """
+        cls.tc_name = cls._get_tc_name()
+        current_date = datetime.now().strftime("%Y%m%d")
+        test_type = getattr(cls, "test_type", "perf")
+        base_output = f"/root/.cache/tests/output/{test_type}/{current_date}"
+        os.makedirs(base_output, exist_ok=True)
+        cls.metrics_data_file = os.path.join(base_output, cls.tc_name)
+        os.makedirs(cls.metrics_data_file, exist_ok=True)
+        # Override env vars so evalscope/dump_metric write to per-case paths.
+        os.environ["METRICS_DATA_FILE"] = cls.metrics_data_file
+        os.environ["SGLANG_TEST_METRICS_OUTPUT"] = os.path.join(
+            cls.metrics_data_file, "metrics"
+        )
+        logger.info(
+            "Per-case output: tc_name=%s metrics_data_file=%s",
+            cls.tc_name,
+            cls.metrics_data_file,
+        )
+
+    @classmethod
+    def _save_metrics_json(cls):
+        """Write per-case ``metrics.json`` from ``dump_metric`` JSONL files.
+
+        Replaces the workflow's stdout-parsing + ``dump_metrics.py`` logic so
+        each case in a suite persists its own metrics snapshot.
+        """
+        if not getattr(cls, "metrics_data_file", None):
+            return
+        metrics = {}
+        baselines = {}
+        pattern = os.path.join(cls.metrics_data_file, "metrics.*.jsonl")
+        for jsonl_path in glob.glob(pattern):
+            try:
+                with open(jsonl_path, "r", encoding="utf-8") as f:
+                    for line in f:
+                        line = line.strip()
+                        if not line:
+                            continue
+                        record = json.loads(line)
+                        name = record.get("metric_name")
+                        value = record.get("value")
+                        if name is None:
+                            continue
+                        if name.endswith("_baseline"):
+                            baselines[name[: -len("_baseline")]] = value
+                        else:
+                            metrics[name] = value
+            except Exception as e:
+                logger.warning("Failed to read %s: %s", jsonl_path, e)
+        out_path = os.path.join(cls.metrics_data_file, "metrics.json")
+        payload = {
+            "test_case": cls.tc_name,
+            "test_type": getattr(cls, "test_type", "accuracy"),
+            "metrics": metrics,
+            "baselines": baselines,
+        }
+        try:
+            with open(out_path, "w", encoding="utf-8") as f:
+                json.dump(payload, f, ensure_ascii=False)
+            logger.info("Saved per-case metrics to %s", out_path)
+        except Exception as e:
+            logger.warning("Failed to write metrics.json: %s", e)
+
+    @classmethod
+    def _backup_plog(cls):
+        """Backup Ascend plog files to a per-case path.
+
+        Replaces the workflow's ``Backup plog`` step so each case in a suite
+        gets its own plog snapshot instead of all cases sharing the suite name.
+        """
+        plog_path = "/root/ascend/log/debug/plog"
+        if not os.path.isdir(plog_path):
+            return
+        tc_name = getattr(cls, "tc_name", None)
+        if not tc_name:
+            return
+        hostname = os.getenv("HOSTNAME", "unknown")
+        target = os.path.join("/root/.cache/tests/logs/plog", tc_name, hostname)
+        os.makedirs(target, exist_ok=True)
+        for name in os.listdir(plog_path):
+            src = os.path.join(plog_path, name)
+            if os.path.isfile(src):
+                try:
+                    shutil.copy2(src, os.path.join(target, name))
+                except Exception as e:
+                    logger.warning("Failed to copy plog %s: %s", name, e)
+        logger.info("Backed up plog to %s", target)
+
     @classmethod
     def setUpClass(cls):
+        cls._setup_per_case_output()
         cls.base_url = DEFAULT_URL_FOR_TEST
         env = os.environ.copy()
         for key, value in env.items():
@@ -937,6 +1050,8 @@ class TestNpuPerformanceTestCaseBase(CustomTestCase):
                 kill_process_tree(cls.process.pid)
             except Exception as e:
                 logger.error(f"Error during tearDown: {e}")
+        cls._save_metrics_json()
+        cls._backup_plog()
 
     @retry()
     def run_throughput(self):
diff --git a/test/registered/npu/accuracy/glm4_7_flash/test_npu_glm4_7_flash_1p_gsm8k.py b/test/registered/npu/accuracy/glm4_7_flash/test_npu_glm4_7_flash_1p_gsm8k.py
index f2de0b3d2..7a24cae35 100644
--- a/test/registered/npu/accuracy/glm4_7_flash/test_npu_glm4_7_flash_1p_gsm8k.py
+++ b/test/registered/npu/accuracy/glm4_7_flash/test_npu_glm4_7_flash_1p_gsm8k.py
@@ -6,12 +6,7 @@ from sglang.test.ascend.e2e.test_npu_accuracy_utils import (
 from sglang.test.ascend.e2e.test_npu_performance_utils import GLM_4_7_FLASH_MODEL_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="accuracy testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-acc-2-npu-a3")
 
 ENVS = {
     "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
diff --git a/test/registered/npu/accuracy/glm5_top64_pruned/test_npu_glm5_top64_pruned_bf16_8p_gsm8k.py b/test/registered/npu/accuracy/glm5_top64_pruned/test_npu_glm5_top64_pruned_bf16_8p_gsm8k.py
index d56b2d5ba..7ae8616ff 100644
--- a/test/registered/npu/accuracy/glm5_top64_pruned/test_npu_glm5_top64_pruned_bf16_8p_gsm8k.py
+++ b/test/registered/npu/accuracy/glm5_top64_pruned/test_npu_glm5_top64_pruned_bf16_8p_gsm8k.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="accuracy testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-acc-16-npu-a3")
 
 ENVS = {
     "SGLANG_SET_CPU_AFFINITY": "1",
@@ -54,7 +49,7 @@ class TestNPUGLM5_Top64_Pruned_GSM8K(TestNpuAccuracyTestCaseBase):
     model = GLM5_TOP64_PRUNED_GSM8K_MODEL_PATH
     envs = ENVS
     other_args = OTHER_ARGS
-    accuracy = 0.50
+    accuracy = 0.48
     datasets = ["gsm8k"]
     generation_config = {
         "max_tokens": 2048,
diff --git a/test/registered/npu/accuracy/moonshotai_moonlight_16b_a3b/test_npu_moonlight_16b_a3b_bf16_1p_gsm8k.py b/test/registered/npu/accuracy/moonshotai_moonlight_16b_a3b/test_npu_moonlight_16b_a3b_bf16_1p_gsm8k.py
index ff490bb46..2a4675e16 100644
--- a/test/registered/npu/accuracy/moonshotai_moonlight_16b_a3b/test_npu_moonlight_16b_a3b_bf16_1p_gsm8k.py
+++ b/test/registered/npu/accuracy/moonshotai_moonlight_16b_a3b/test_npu_moonlight_16b_a3b_bf16_1p_gsm8k.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="accuracy testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-acc-2-npu-a3")
 
 MODEL_ENVS = {
     "SGLANG_SET_CPU_AFFINITY": "1",
diff --git a/test/registered/npu/accuracy/qwen3_5_9b/test_npu_qwen3_5_9b_bf16_1p_gsm8k.py b/test/registered/npu/accuracy/qwen3_5_9b/test_npu_qwen3_5_9b_bf16_1p_gsm8k.py
index 7b49dd52e..c997de8bf 100644
--- a/test/registered/npu/accuracy/qwen3_5_9b/test_npu_qwen3_5_9b_bf16_1p_gsm8k.py
+++ b/test/registered/npu/accuracy/qwen3_5_9b/test_npu_qwen3_5_9b_bf16_1p_gsm8k.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="stage-b-test-2-npu-a3",
-    nightly=True,
-    disabled="accuracy testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-acc-2-npu-a3")
 
 QWEN3_5_9B_ENVS = {
     "SGLANG_SET_CPU_AFFINITY": "1",
diff --git a/test/registered/npu/accuracy/qwen3_vl_30b_a3b/test_npu_qwen3_vl_30b_a3b_bf16_2p_gsm8k.py b/test/registered/npu/accuracy/qwen3_vl_30b_a3b/test_npu_qwen3_vl_30b_a3b_bf16_2p_gsm8k.py
index ffaf7e9de..0d9e06817 100644
--- a/test/registered/npu/accuracy/qwen3_vl_30b_a3b/test_npu_qwen3_vl_30b_a3b_bf16_2p_gsm8k.py
+++ b/test/registered/npu/accuracy/qwen3_vl_30b_a3b/test_npu_qwen3_vl_30b_a3b_bf16_2p_gsm8k.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="stage-b-test-4-npu-a3",
-    nightly=True,
-    disabled="accuracy testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-acc-4-npu-a3")
 
 QWEN3_VL_30B_A3B_ENVS = {
     "SGLANG_SET_CPU_AFFINITY": "1",
diff --git a/test/registered/npu/accuracy/qwen3_vl_8b/test_npu_qwen3_vl_8b_bf16_2p_gsm8k.py b/test/registered/npu/accuracy/qwen3_vl_8b/test_npu_qwen3_vl_8b_bf16_2p_gsm8k.py
index 6d44fc382..ffca0f1ed 100644
--- a/test/registered/npu/accuracy/qwen3_vl_8b/test_npu_qwen3_vl_8b_bf16_2p_gsm8k.py
+++ b/test/registered/npu/accuracy/qwen3_vl_8b/test_npu_qwen3_vl_8b_bf16_2p_gsm8k.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="stage-b-test-4-npu-a3",
-    nightly=True,
-    disabled="accuracy testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-acc-4-npu-a3")
 
 QWEN3_VL_8B_ENVS = {
     "SGLANG_SET_CPU_AFFINITY": "1",
diff --git a/test/registered/npu/accuracy/qwen3_vl_8b_thinking/test_npu_qwen3_vl_8b_thinking_1p_mmmu.py b/test/registered/npu/accuracy/qwen3_vl_8b_thinking/test_npu_qwen3_vl_8b_thinking_1p_mmmu.py
index 2f9ce2f8d..1cfefbc1f 100644
--- a/test/registered/npu/accuracy/qwen3_vl_8b_thinking/test_npu_qwen3_vl_8b_thinking_1p_mmmu.py
+++ b/test/registered/npu/accuracy/qwen3_vl_8b_thinking/test_npu_qwen3_vl_8b_thinking_1p_mmmu.py
@@ -9,12 +9,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-2-npu-a3")
 
 _is_pr_pipeline = os.environ.get("GITHUB_EVENT_NAME") == "pull_request"
 
diff --git a/test/registered/npu/basic_function/HiCache/test_npu_hicache_mha.py b/test/registered/npu/basic_function/HiCache/test_npu_hicache_mha.py
index b83fdaf70..99b843ed2 100644
--- a/test/registered/npu/basic_function/HiCache/test_npu_hicache_mha.py
+++ b/test/registered/npu/basic_function/HiCache/test_npu_hicache_mha.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/HiCache/test_npu_hicache_mla.py b/test/registered/npu/basic_function/HiCache/test_npu_hicache_mla.py
index 9448071ed..496a38373 100644
--- a/test/registered/npu/basic_function/HiCache/test_npu_hicache_mla.py
+++ b/test/registered/npu/basic_function/HiCache/test_npu_hicache_mla.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-4-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/backends/test_npu_sampling_backend.py b/test/registered/npu/basic_function/backends/test_npu_sampling_backend.py
index 8b278d3f5..df93ae543 100644
--- a/test/registered/npu/basic_function/backends/test_npu_sampling_backend.py
+++ b/test/registered/npu/basic_function/backends/test_npu_sampling_backend.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/dllm/test_npu_llada2_mini.py b/test/registered/npu/basic_function/dllm/test_npu_llada2_mini.py
index 4d55a335a..dc55d5565 100644
--- a/test/registered/npu/basic_function/dllm/test_npu_llada2_mini.py
+++ b/test/registered/npu/basic_function/dllm/test_npu_llada2_mini.py
@@ -10,7 +10,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-4-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/dp_attn/test_npu_dp_attention.py b/test/registered/npu/basic_function/dp_attn/test_npu_dp_attention.py
index a93a94312..a830ea2b6 100644
--- a/test/registered/npu/basic_function/dp_attn/test_npu_dp_attention.py
+++ b/test/registered/npu/basic_function/dp_attn/test_npu_dp_attention.py
@@ -23,7 +23,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-4-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/optimization_debug/test_npu_compile_graph_tp1_bf16.py b/test/registered/npu/basic_function/optimization_debug/test_npu_compile_graph_tp1_bf16.py
index 9255b5337..05cad7f0e 100644
--- a/test/registered/npu/basic_function/optimization_debug/test_npu_compile_graph_tp1_bf16.py
+++ b/test/registered/npu/basic_function/optimization_debug/test_npu_compile_graph_tp1_bf16.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/optimization_debug/test_npu_piecewise_graph_prefill.py b/test/registered/npu/basic_function/optimization_debug/test_npu_piecewise_graph_prefill.py
index b4b620c9f..c3230817b 100644
--- a/test/registered/npu/basic_function/optimization_debug/test_npu_piecewise_graph_prefill.py
+++ b/test/registered/npu/basic_function/optimization_debug/test_npu_piecewise_graph_prefill.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     run_bench_one_batch,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/parallel_strategy/data_parallelism/test_npu_load_balance_method.py b/test/registered/npu/basic_function/parallel_strategy/data_parallelism/test_npu_load_balance_method.py
index 7f2a60492..cf347e0be 100644
--- a/test/registered/npu/basic_function/parallel_strategy/data_parallelism/test_npu_load_balance_method.py
+++ b/test/registered/npu/basic_function/parallel_strategy/data_parallelism/test_npu_load_balance_method.py
@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=700, suite="stage-b-test-16-npu-a3", nightly=False)
+register_npu_ci(est_time=700, suite="base-b-test-16-npu-a3")
 register_npu_ci(est_time=700, suite="nightly-16-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/parallel_strategy/data_parallelism/test_npu_load_balance_method_pd_disaggregation.py b/test/registered/npu/basic_function/parallel_strategy/data_parallelism/test_npu_load_balance_method_pd_disaggregation.py
index 6a4075cd4..730ae28ec 100644
--- a/test/registered/npu/basic_function/parallel_strategy/data_parallelism/test_npu_load_balance_method_pd_disaggregation.py
+++ b/test/registered/npu/basic_function/parallel_strategy/data_parallelism/test_npu_load_balance_method_pd_disaggregation.py
@@ -23,7 +23,7 @@ from sglang.test.test_utils import (
     popen_launch_pd_server,
 )
 
-register_npu_ci(est_time=3600, suite="stage-b-test-16-npu-a3", nightly=False)
+register_npu_ci(est_time=3600, suite="base-b-test-16-npu-a3")
 register_npu_ci(est_time=3600, suite="nightly-16-npu-a3", nightly=True)
 
 load_balance_method_options = [
diff --git a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_deepep.py b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_deepep.py
index 949701643..e7b158622 100644
--- a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_deepep.py
+++ b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_deepep.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-16-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-16-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-16-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_eplb_min_rebalancing_utilization_threshold.py b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_eplb_min_rebalancing_utilization_threshold.py
index 057a09b51..1c23e7df1 100644
--- a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_eplb_min_rebalancing_utilization_threshold.py
+++ b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_eplb_min_rebalancing_utilization_threshold.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-8-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-8-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-8-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_expert_distribution_recorder_mode.py b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_expert_distribution_recorder_mode.py
index 27497235b..8d96cc9b4 100644
--- a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_expert_distribution_recorder_mode.py
+++ b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_expert_distribution_recorder_mode.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-2-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_moe_dense_tp_size.py b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_moe_dense_tp_size.py
index 1e1364a7c..cdc39ff86 100644
--- a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_moe_dense_tp_size.py
+++ b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_moe_dense_tp_size.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-2-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_moe_runner_backend.py b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_moe_runner_backend.py
index 24f6c6ca9..a2db0c6f7 100644
--- a/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_moe_runner_backend.py
+++ b/test/registered/npu/basic_function/parallel_strategy/expert_parallelism/test_npu_moe_runner_backend.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-2-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/pd_disaggregation/test_npu_pd_disaggregation.py b/test/registered/npu/basic_function/pd_disaggregation/test_npu_pd_disaggregation.py
index 7585c27c8..e3e339375 100644
--- a/test/registered/npu/basic_function/pd_disaggregation/test_npu_pd_disaggregation.py
+++ b/test/registered/npu/basic_function/pd_disaggregation/test_npu_pd_disaggregation.py
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
     popen_with_error_check,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-16-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-16-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-16-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/quant/test_npu_autoround_dense.py b/test/registered/npu/basic_function/quant/test_npu_autoround_dense.py
index f684e5990..97a8249ec 100644
--- a/test/registered/npu/basic_function/quant/test_npu_autoround_dense.py
+++ b/test/registered/npu/basic_function/quant/test_npu_autoround_dense.py
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 logger = logging.getLogger(__name__)
diff --git a/test/registered/npu/basic_function/quant/test_npu_autoround_moe.py b/test/registered/npu/basic_function/quant/test_npu_autoround_moe.py
index a9641ec69..1c831a938 100644
--- a/test/registered/npu/basic_function/quant/test_npu_autoround_moe.py
+++ b/test/registered/npu/basic_function/quant/test_npu_autoround_moe.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 logger = logging.getLogger(__name__)
diff --git a/test/registered/npu/basic_function/quant/test_npu_gptq_moe.py b/test/registered/npu/basic_function/quant/test_npu_gptq_moe.py
index b305f2872..bf9d4a296 100644
--- a/test/registered/npu/basic_function/quant/test_npu_gptq_moe.py
+++ b/test/registered/npu/basic_function/quant/test_npu_gptq_moe.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 logger = logging.getLogger(__name__)
diff --git a/test/registered/npu/basic_function/quant/test_npu_w4a4_quantization.py b/test/registered/npu/basic_function/quant/test_npu_w4a4_quantization.py
index 3db391e6c..f3743bcb0 100644
--- a/test/registered/npu/basic_function/quant/test_npu_w4a4_quantization.py
+++ b/test/registered/npu/basic_function/quant/test_npu_w4a4_quantization.py
@@ -13,7 +13,7 @@ from sglang.test.ascend.test_ascend_utils import (
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="stage-b-test-4-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/runtime_opts/test_npu_mla_fia_w8a8int8.py b/test/registered/npu/basic_function/runtime_opts/test_npu_mla_fia_w8a8int8.py
index 75d4bcc1a..0fb4236be 100644
--- a/test/registered/npu/basic_function/runtime_opts/test_npu_mla_fia_w8a8int8.py
+++ b/test/registered/npu/basic_function/runtime_opts/test_npu_mla_fia_w8a8int8.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-2-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/runtime_opts/test_npu_mla_w8a8int8.py b/test/registered/npu/basic_function/runtime_opts/test_npu_mla_w8a8int8.py
index 94c76b258..50dcd88fa 100644
--- a/test/registered/npu/basic_function/runtime_opts/test_npu_mla_w8a8int8.py
+++ b/test/registered/npu/basic_function/runtime_opts/test_npu_mla_w8a8int8.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-4-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/runtime_opts/test_npu_tp1_bf16.py b/test/registered/npu/basic_function/runtime_opts/test_npu_tp1_bf16.py
index e9740342e..1556c62d7 100644
--- a/test/registered/npu/basic_function/runtime_opts/test_npu_tp1_bf16.py
+++ b/test/registered/npu/basic_function/runtime_opts/test_npu_tp1_bf16.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/runtime_opts/test_npu_tp2_bf16.py b/test/registered/npu/basic_function/runtime_opts/test_npu_tp2_bf16.py
index 2ec3d16dd..5a88ad3b0 100644
--- a/test/registered/npu/basic_function/runtime_opts/test_npu_tp2_bf16.py
+++ b/test/registered/npu/basic_function/runtime_opts/test_npu_tp2_bf16.py
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-2-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/runtime_opts/test_npu_tp2_fia_bf16.py b/test/registered/npu/basic_function/runtime_opts/test_npu_tp2_fia_bf16.py
index 821f09abe..bf2ff3361 100644
--- a/test/registered/npu/basic_function/runtime_opts/test_npu_tp2_fia_bf16.py
+++ b/test/registered/npu/basic_function/runtime_opts/test_npu_tp2_fia_bf16.py
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-2-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/runtime_opts/test_npu_tp4_bf16.py b/test/registered/npu/basic_function/runtime_opts/test_npu_tp4_bf16.py
index b94887783..f9842d10b 100644
--- a/test/registered/npu/basic_function/runtime_opts/test_npu_tp4_bf16.py
+++ b/test/registered/npu/basic_function/runtime_opts/test_npu_tp4_bf16.py
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-4-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)
 
 TEST_MODEL_MATRIX = {
diff --git a/test/registered/npu/basic_function/speculative_inference/test_npu_basic_sanity_eagle3.py b/test/registered/npu/basic_function/speculative_inference/test_npu_basic_sanity_eagle3.py
index 7254361fb..979a01bee 100644
--- a/test/registered/npu/basic_function/speculative_inference/test_npu_basic_sanity_eagle3.py
+++ b/test/registered/npu/basic_function/speculative_inference/test_npu_basic_sanity_eagle3.py
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=200, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=200, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=200, suite="nightly-1-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/speculative_inference/test_npu_eagle3.py b/test/registered/npu/basic_function/speculative_inference/test_npu_eagle3.py
index 5c56818c3..21c914aeb 100644
--- a/test/registered/npu/basic_function/speculative_inference/test_npu_eagle3.py
+++ b/test/registered/npu/basic_function/speculative_inference/test_npu_eagle3.py
@@ -9,7 +9,7 @@ from sglang.test.ascend.test_ascend_utils import (
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_attention_mode.py b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_attention_mode.py
index c6cec152a..cfc9dddec 100644
--- a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_attention_mode.py
+++ b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_attention_mode.py
@@ -17,11 +17,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(
-    est_time=400,
-    suite="stage-b-test-4-npu-a3",
-    nightly=False,
-)
+register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3")
 register_npu_ci(
     est_time=400,
     suite="nightly-4-npu-a3",
diff --git a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_draft_attention_backend.py b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_draft_attention_backend.py
index c61021bb9..485e54b4f 100644
--- a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_draft_attention_backend.py
+++ b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_draft_attention_backend.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-16-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-16-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-16-npu-a3", nightly=True)
 
 MODEL_PATH = DEEPSEEK_R1_0528_W4A8_PER_CHANNEL_WEIGHTS_PATH
diff --git a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_moe_a2a_backend.py b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_moe_a2a_backend.py
index e02518785..56573a783 100644
--- a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_moe_a2a_backend.py
+++ b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_moe_a2a_backend.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-16-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-16-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-16-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_multi_npu.py b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_multi_npu.py
index e9185a06a..8a9564b2d 100644
--- a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_multi_npu.py
+++ b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_multi_npu.py
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=400, suite="stage-b-test-4-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="base-b-test-4-npu-a3")
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)
 
 _ASCEND_BACKEND = "ascend"
diff --git a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_token_map.py b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_token_map.py
index 8d0400b28..3cde909da 100644
--- a/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_token_map.py
+++ b/test/registered/npu/basic_function/speculative_inference/test_npu_speculative_token_map.py
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_npu_ci(est_time=300, suite="stage-b-test-4-npu-a3", nightly=False)
+register_npu_ci(est_time=300, suite="base-b-test-4-npu-a3")
 register_npu_ci(est_time=300, suite="nightly-4-npu-a3", nightly=True)
 
 
diff --git a/test/registered/npu/performance/deepseek_v4_flash/test_npu_deepseek_v4_flash_w8a8_8p_in8k_out1k_50ms.py b/test/registered/npu/performance/deepseek_v4_flash/test_npu_deepseek_v4_flash_w8a8_8p_in8k_out1k_50ms.py
index 62fbef9a7..e8ae7403d 100644
--- a/test/registered/npu/performance/deepseek_v4_flash/test_npu_deepseek_v4_flash_w8a8_8p_in8k_out1k_50ms.py
+++ b/test/registered/npu/performance/deepseek_v4_flash/test_npu_deepseek_v4_flash_w8a8_8p_in8k_out1k_50ms.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-16-npu-a3")
 
 # Environment variables for DSV4-Flash single-node PD-mix deployment.
 DEEPSEEK_V4_FLASH_W8A8_8P_ENVS = {
diff --git a/test/registered/npu/performance/kimi_k2_6/test_npu_kimi_k2_6_w4a8_8p_in3k5_out1k5_20ms.py b/test/registered/npu/performance/kimi_k2_6/test_npu_kimi_k2_6_w4a8_8p_in3k5_out1k5_20ms.py
index 43c7cc761..8ad487567 100644
--- a/test/registered/npu/performance/kimi_k2_6/test_npu_kimi_k2_6_w4a8_8p_in3k5_out1k5_20ms.py
+++ b/test/registered/npu/performance/kimi_k2_6/test_npu_kimi_k2_6_w4a8_8p_in3k5_out1k5_20ms.py
@@ -10,12 +10,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=1800,
-    suite="full-16-npu-a3",
-    nightly=True,
-    disabled="Currently it is executed by the npu performance workflow.",
-)
+register_npu_ci(est_time=1800, suite="base-c-test-perf-16-npu-a3")
 
 KIMI_K2_6_ENVS = {
     "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
diff --git a/test/registered/npu/performance/minimax_m2_5/test_npu_minimax_m2_5_w8a8_4p_in64k_out1k_prefix90_50ms.py b/test/registered/npu/performance/minimax_m2_5/test_npu_minimax_m2_5_w8a8_4p_in64k_out1k_prefix90_50ms.py
index a1d35c737..d37cf0f55 100644
--- a/test/registered/npu/performance/minimax_m2_5/test_npu_minimax_m2_5_w8a8_4p_in64k_out1k_prefix90_50ms.py
+++ b/test/registered/npu/performance/minimax_m2_5/test_npu_minimax_m2_5_w8a8_4p_in64k_out1k_prefix90_50ms.py
@@ -9,12 +9,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="npu-performance",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-8-npu-a3")
 
 MINIMAX_M2_5_W8A8_4P_IN64K_OUT1K_PREFIX90_ENVS = {
     "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
diff --git a/test/registered/npu/performance/qwen3-8b/test_npu_qwen3_8b_w8a8_1p_in3k5_out1k5_50ms.py b/test/registered/npu/performance/qwen3-8b/test_npu_qwen3_8b_w8a8_1p_in3k5_out1k5_50ms.py
index 4509463c4..125c55ac5 100644
--- a/test/registered/npu/performance/qwen3-8b/test_npu_qwen3_8b_w8a8_1p_in3k5_out1k5_50ms.py
+++ b/test/registered/npu/performance/qwen3-8b/test_npu_qwen3_8b_w8a8_1p_in3k5_out1k5_50ms.py
@@ -9,12 +9,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-2-npu-a3")
 
 QWEN3_8B_ENVS = {
     "SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT": "600",
diff --git a/test/registered/npu/performance/qwen3_235b_a22b/test_npu_qwen3_235b_w8a8_8p_in3k5_out1k5_50ms.py b/test/registered/npu/performance/qwen3_235b_a22b/test_npu_qwen3_235b_w8a8_8p_in3k5_out1k5_50ms.py
index 74335bafc..d115b37ff 100644
--- a/test/registered/npu/performance/qwen3_235b_a22b/test_npu_qwen3_235b_w8a8_8p_in3k5_out1k5_50ms.py
+++ b/test/registered/npu/performance/qwen3_235b_a22b/test_npu_qwen3_235b_w8a8_8p_in3k5_out1k5_50ms.py
@@ -9,12 +9,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-16-npu-a3")
 
 QWEN3_235B_ENVS = {
     "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
diff --git a/test/registered/npu/performance/qwen3_30b_a3b/test_npu_qwen3_30b_w8a8_1p_in3k5_out1k5_50ms.py b/test/registered/npu/performance/qwen3_30b_a3b/test_npu_qwen3_30b_w8a8_1p_in3k5_out1k5_50ms.py
index 3bf536ac5..50ebda549 100644
--- a/test/registered/npu/performance/qwen3_30b_a3b/test_npu_qwen3_30b_w8a8_1p_in3k5_out1k5_50ms.py
+++ b/test/registered/npu/performance/qwen3_30b_a3b/test_npu_qwen3_30b_w8a8_1p_in3k5_out1k5_50ms.py
@@ -9,12 +9,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-2-npu-a3")
 
 QWEN3_30B_A3B_ENVS = {
     "ASCEND_LAUNCH_BLOCKING": "0",
diff --git a/test/registered/npu/performance/qwen3_32b/test_npu_qwen3_32b_w8a8_2p_in3k5_out1k5_50ms.py b/test/registered/npu/performance/qwen3_32b/test_npu_qwen3_32b_w8a8_2p_in3k5_out1k5_50ms.py
index 89cf2a304..7752ee5b5 100644
--- a/test/registered/npu/performance/qwen3_32b/test_npu_qwen3_32b_w8a8_2p_in3k5_out1k5_50ms.py
+++ b/test/registered/npu/performance/qwen3_32b/test_npu_qwen3_32b_w8a8_2p_in3k5_out1k5_50ms.py
@@ -9,12 +9,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-4-npu-a3")
 
 QWEN3_32B_ENVS = {
     "SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT": "600",
diff --git a/test/registered/npu/performance/qwen3_5_397b/test_npu_qwen3_5_397b_w4a8_8p_in3k5_out1k5_50ms.py b/test/registered/npu/performance/qwen3_5_397b/test_npu_qwen3_5_397b_w4a8_8p_in3k5_out1k5_50ms.py
index 68d5d75ef..c77ef4810 100644
--- a/test/registered/npu/performance/qwen3_5_397b/test_npu_qwen3_5_397b_w4a8_8p_in3k5_out1k5_50ms.py
+++ b/test/registered/npu/performance/qwen3_5_397b/test_npu_qwen3_5_397b_w4a8_8p_in3k5_out1k5_50ms.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="nightly-16-npu-a3",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-16-npu-a3")
 
 QWEN3_5_397B_A17B_ENVS = {
     "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
diff --git a/test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_1p_in1024x1024_30_out1024_50ms.py b/test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_1p_in1024x1024_30_out1024_50ms.py
index fd501d5ff..0bd8aab8d 100644
--- a/test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_1p_in1024x1024_30_out1024_50ms.py
+++ b/test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_1p_in1024x1024_30_out1024_50ms.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="full-2-npu-a3",
-    nightly=True,
-    disabled="performance case",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-2-npu-a3")
 
 QWEN3_6_27B_1024_ENVS = {
     "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
diff --git a/test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_w8a8_1p_in64k_out1k_50ms.py b/test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_w8a8_1p_in64k_out1k_50ms.py
index 68f8f341c..14b869e4c 100644
--- a/test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_w8a8_1p_in64k_out1k_50ms.py
+++ b/test/registered/npu/performance/qwen3_6_27b/test_npu_qwen3_6_27b_w8a8_1p_in64k_out1k_50ms.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-2-npu-a3")
 
 QWEN3_6_27B_64K_1K_ENVS = {
     "STREAMS_PER_DEVICE": "32",
diff --git a/test/registered/npu/performance/qwen3_6_35b_a3b/test_npu_qwen3_6_35b_a3b_1p_in64k_out1k_prefix90_50ms.py b/test/registered/npu/performance/qwen3_6_35b_a3b/test_npu_qwen3_6_35b_a3b_1p_in64k_out1k_prefix90_50ms.py
index 371a9cf43..cd191e7a4 100644
--- a/test/registered/npu/performance/qwen3_6_35b_a3b/test_npu_qwen3_6_35b_a3b_1p_in64k_out1k_prefix90_50ms.py
+++ b/test/registered/npu/performance/qwen3_6_35b_a3b/test_npu_qwen3_6_35b_a3b_1p_in64k_out1k_prefix90_50ms.py
@@ -8,12 +8,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-2-npu-a3")
 
 QWEN3_6_35B_A3B_64K_PREFIX_ENVS = {
     "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
diff --git a/test/registered/npu/performance/qwen3_next_80b_a3b_instruct/test_npu_qwen3_next_80b_w8a8_2p_in6k_out1k5_bs16.py b/test/registered/npu/performance/qwen3_next_80b_a3b_instruct/test_npu_qwen3_next_80b_w8a8_2p_in6k_out1k5_bs16.py
index 1952c1ef8..155ed324a 100644
--- a/test/registered/npu/performance/qwen3_next_80b_a3b_instruct/test_npu_qwen3_next_80b_w8a8_2p_in6k_out1k5_bs16.py
+++ b/test/registered/npu/performance/qwen3_next_80b_a3b_instruct/test_npu_qwen3_next_80b_w8a8_2p_in6k_out1k5_bs16.py
@@ -7,12 +7,7 @@ from sglang.test.ascend.e2e.test_npu_performance_utils import (
 )
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(
-    est_time=3600,
-    suite="",
-    nightly=True,
-    disabled="performance testcase",
-)
+register_npu_ci(est_time=3600, suite="base-c-test-perf-4-npu-a3")
 
 QWEN3_NEXT_80B_A3B_ENVS = {
     "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
diff --git a/test/registered/unit/layers/attention/test_vision_backend_selection.py b/test/registered/unit/layers/attention/test_vision_backend_selection.py
index b7e6296fa..3ed5a4d64 100644
--- a/test/registered/unit/layers/attention/test_vision_backend_selection.py
+++ b/test/registered/unit/layers/attention/test_vision_backend_selection.py
@@ -11,7 +11,7 @@ from sglang.srt.layers.attention import vision
 from sglang.test.ci.ci_register import register_cpu_ci, register_npu_ci
 
 register_cpu_ci(est_time=2, suite="base-a-test-cpu")
-register_npu_ci(est_time=2, suite="stage-b-test-1-npu-a3")
+register_npu_ci(est_time=2, suite="base-b-test-1-npu-a3")
 register_npu_ci(est_time=2, suite="nightly-1-npu-a3", nightly=True)
 
 
diff --git a/test/registered/unit/npu/attention/test_npu_ascend_backend.py b/test/registered/unit/npu/attention/test_npu_ascend_backend.py
index 867a4a38a..ddc6e9093 100644
--- a/test/registered/unit/npu/attention/test_npu_ascend_backend.py
+++ b/test/registered/unit/npu/attention/test_npu_ascend_backend.py
@@ -12,7 +12,7 @@ import torch
 
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(est_time=5, suite="stage-a-unit-test-npu")
+register_npu_ci(est_time=5, suite="base-a-test-1-npu-a2")
 
 # Mock NPU-only modules before importing the source module.
 for _ in (
diff --git a/test/registered/unit/npu/attention/test_npu_ascend_dsv4_backend.py b/test/registered/unit/npu/attention/test_npu_ascend_dsv4_backend.py
index 5cf5f1e16..c244c1d24 100644
--- a/test/registered/unit/npu/attention/test_npu_ascend_dsv4_backend.py
+++ b/test/registered/unit/npu/attention/test_npu_ascend_dsv4_backend.py
@@ -12,7 +12,7 @@ import torch
 
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(est_time=4, suite="stage-a-unit-test-npu")
+register_npu_ci(est_time=4, suite="base-a-test-1-npu-a2")
 
 for mod in (
     "torch_npu",
diff --git a/test/run_suite.py b/test/run_suite.py
index 1af0d5b20..b4ac59623 100644
--- a/test/run_suite.py
+++ b/test/run_suite.py
@@ -93,13 +93,20 @@ PER_COMMIT_SUITES = {
         "extra-b-test-8-gpu-h200",
     ],
     HWBackend.NPU: [
-        "base-a-test-1-gpu-small",
-        "stage-a-unit-test-npu",
-        "stage-b-test-1-npu-a3",
-        "stage-b-test-2-npu-a3",
-        "stage-b-test-4-npu-a3",
-        "stage-b-test-8-npu-a3",
-        "stage-b-test-16-npu-a3",
+        "base-a-test-1-npu-a2",
+        "base-b-test-1-npu-a3",
+        "base-b-test-2-npu-a3",
+        "base-b-test-4-npu-a3",
+        "base-b-test-8-npu-a3",
+        "base-b-test-16-npu-a3",
+        "base-c-test-acc-2-npu-a3",
+        "base-c-test-acc-4-npu-a3",
+        "base-c-test-acc-8-npu-a3",
+        "base-c-test-acc-16-npu-a3",
+        "base-c-test-perf-2-npu-a3",
+        "base-c-test-perf-4-npu-a3",
+        "base-c-test-perf-8-npu-a3",
+        "base-c-test-perf-16-npu-a3",
     ],
     HWBackend.XPU: [
         "stage-a-test-1-gpu-xpu",