From e67b95d66b092fa36f666fc2a575e76f45464441 Mon Sep 17 00:00:00 2001
From: Cherry_ming <136634645@qq.com>
Date: Wed, 1 Apr 2026 19:56:31 +0800
Subject: [PATCH] [NPU]Add a full test pipeline on NPU, resolve issues in the
 NPU test architecture (#20751)

---
 .github/workflows/full-test-npu.yml           | 355 ++++++++++++++++++
 .github/workflows/nightly-test-npu.yml        | 160 ++++++--
 .github/workflows/pr-test-npu.yml             | 110 ++++--
 .../workflows/release-docker-npu-nightly.yml  |   2 +-
 python/pyproject_npu.toml                     |   2 +
 .../sglang/test/ascend/test_ascend_utils.py   |  18 +-
 .../HiCache/test_npu_hicache_mha.py}          |   4 +
 .../HiCache/test_npu_hicache_mla.py}          |   4 +
 .../backends/test_npu_sampling_backend.py}    |   4 +
 .../dllm/test_npu_llada2_mini.py}             |   4 +
 .../test_npu_compile_graph_tp1_bf16.py}       |   4 +
 .../test_npu_graph_tp1_bf16.py}               |   4 +
 .../test_npu_graph_tp2_bf16.py}               |   4 +
 .../test_npu_piecewise_graph_prefill.py}      |   4 +
 .../expert_parallelism/test_npu_deepep.py}    |   4 +
 .../quant/test_npu_autoround_dense.py}        |   7 +-
 .../quant/test_npu_autoround_moe.py}          |   9 +-
 .../quant/test_npu_gptq_moe.py}               |   9 +-
 .../quant/test_npu_w4a4_quantization.py}      |   4 +
 .../quant/test_npu_w8a8_quantization.py}      |   4 +
 .../test_npu_mla_fia_w8a8int8.py}             |   4 +
 .../runtime_opts/test_npu_mla_w8a8int8.py}    |   4 +
 .../runtime_opts/test_npu_tp1_bf16.py}        |   4 +
 .../runtime_opts/test_npu_tp2_bf16.py}        |   4 +
 .../runtime_opts/test_npu_tp2_fia_bf16.py}    |   4 +
 .../runtime_opts/test_npu_tp4_bf16.py}        |   4 +
 .../test_npu_bge_large_en_v1_5.py             |   2 +-
 .../ascend/llm_models/test_npu_afm_4_5b.py    |   2 +-
 .../llm_models/test_npu_c4ai_command_r_v01.py |   2 +-
 .../ascend/llm_models/test_npu_exaone_3.py    |   2 +-
 .../test_npu_granite_3_0_3b_a800m.py          |   2 +-
 .../llm_models/test_npu_granite_3_1_8b.py     |   2 +-
 .../ascend/llm_models/test_npu_grok_2.py      |   2 +-
 .../ascend/llm_models/test_npu_ling_lite.py   |   2 +-
 .../ascend/llm_models/test_npu_mimo_7b_rl.py  |   2 +-
 .../llm_models/test_npu_persimmon_8b_chat.py  |   2 +-
 .../ascend/llm_models/test_npu_smollm_1_7b.py |   2 +-
 .../llm_models/test_npu_stablelm_2_1_6b.py    |   2 +-
 .../test_npu_gemma_2_27b_v0_2.py              |   2 +-
 .../test_npu_internlm2_7b_reward.py           |   4 +-
 .../test_npu_llama_3_1_8b_v0_2.py             |   2 +-
 .../test_npu_qwen3_vl_4b_instruct.py          |   2 +-
 test/run_suite.py                             |   5 +
 43 files changed, 673 insertions(+), 106 deletions(-)
 create mode 100644 .github/workflows/full-test-npu.yml
 rename test/{srt/ascend/test_ascend_hicache_mha.py => registered/ascend/basic_function/HiCache/test_npu_hicache_mha.py} (91%)
 rename test/{srt/ascend/test_ascend_hicache_mla.py => registered/ascend/basic_function/HiCache/test_npu_hicache_mla.py} (91%)
 rename test/{srt/ascend/test_ascend_sampling_backend.py => registered/ascend/basic_function/backends/test_npu_sampling_backend.py} (92%)
 rename test/{srt/ascend/test_llada2_mini_ascend.py => registered/ascend/basic_function/dllm/test_npu_llada2_mini.py} (92%)
 rename test/{srt/ascend/test_ascend_compile_graph_tp1_bf16.py => registered/ascend/basic_function/optimization_debug/test_npu_compile_graph_tp1_bf16.py} (92%)
 rename test/{srt/ascend/test_ascend_graph_tp1_bf16.py => registered/ascend/basic_function/optimization_debug/test_npu_graph_tp1_bf16.py} (91%)
 rename test/{srt/ascend/test_ascend_graph_tp2_bf16.py => registered/ascend/basic_function/optimization_debug/test_npu_graph_tp2_bf16.py} (91%)
 rename test/{srt/ascend/test_ascend_piecewise_graph_prefill.py => registered/ascend/basic_function/optimization_debug/test_npu_piecewise_graph_prefill.py} (92%)
 rename test/{srt/ascend/test_ascend_deepep.py => registered/ascend/basic_function/parallel_strategy/expert_parallelism/test_npu_deepep.py} (93%)
 rename test/{srt/ascend/test_ascend_autoround_dense.py => registered/ascend/basic_function/quant/test_npu_autoround_dense.py} (86%)
 rename test/{srt/ascend/test_ascend_autoround_moe.py => registered/ascend/basic_function/quant/test_npu_autoround_moe.py} (85%)
 rename test/{srt/ascend/test_ascend_gptq_moe.py => registered/ascend/basic_function/quant/test_npu_gptq_moe.py} (86%)
 rename test/{srt/ascend/test_ascend_w4a4_quantization.py => registered/ascend/basic_function/quant/test_npu_w4a4_quantization.py} (93%)
 rename test/{srt/ascend/test_ascend_w8a8_quantization.py => registered/ascend/basic_function/quant/test_npu_w8a8_quantization.py} (93%)
 rename test/{srt/ascend/test_ascend_mla_fia_w8a8int8.py => registered/ascend/basic_function/runtime_opts/test_npu_mla_fia_w8a8int8.py} (92%)
 rename test/{srt/ascend/test_ascend_mla_w8a8int8.py => registered/ascend/basic_function/runtime_opts/test_npu_mla_w8a8int8.py} (91%)
 rename test/{srt/ascend/test_ascend_tp1_bf16.py => registered/ascend/basic_function/runtime_opts/test_npu_tp1_bf16.py} (91%)
 rename test/{srt/ascend/test_ascend_tp2_bf16.py => registered/ascend/basic_function/runtime_opts/test_npu_tp2_bf16.py} (91%)
 rename test/{srt/ascend/test_ascend_tp2_fia_bf16.py => registered/ascend/basic_function/runtime_opts/test_npu_tp2_fia_bf16.py} (92%)
 rename test/{srt/ascend/test_ascend_tp4_bf16.py => registered/ascend/basic_function/runtime_opts/test_npu_tp4_bf16.py} (91%)

diff --git a/.github/workflows/full-test-npu.yml b/.github/workflows/full-test-npu.yml
new file mode 100644
index 000000000..47355f2c2
--- /dev/null
+++ b/.github/workflows/full-test-npu.yml
@@ -0,0 +1,355 @@
+name: Full Test (NPU)
+
+on:
+#  pull_request:
+#    branches:
+#      - main
+#    paths:
+#      - ".github/workflows/full-test-npu.yml"
+  workflow_dispatch:
+    inputs:
+      ref:
+        description: 'Git ref (branch, tag, or SHA) to test. If not provided, uses the default branch.'
+        required: false
+        type: string
+        default: ''
+      job_filter:
+        description: 'Select which job to run (leave empty or "all" to run all jobs)'
+        required: false
+        type: string
+        default: 'all'
+      image_a3:
+        description: 'The a3 running docker image of the test task.'
+        required: false
+        type: string
+        default: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11'
+      skip_install_flag:
+        description: 'Indicates whether to skip the installation of sglang, defaulting to false.'
+        required: false
+        type: string
+        default: 'false'
+
+concurrency:
+  group: full-test-npu-${{ inputs.ref || github.ref }}
+  cancel-in-progress: ${{ github.event_name != 'workflow_call' }}
+
+jobs:
+  set-image-config:
+    runs-on: ubuntu-latest
+    outputs:
+      ref: ${{ steps.set-vars.outputs.ref }}
+      job_filter: ${{ steps.set-vars.outputs.job_filter }}
+      image_a3: ${{ steps.set-vars.outputs.image_a3 }}
+      skip_install_flag: ${{ steps.set-vars.outputs.skip_install_flag }}
+    steps:
+      # When triggered by PR, no inputs parameters are used. The latest community code is tested by default.
+      - name: Set image config
+        id: set-vars
+        run: |
+          if [ -z "${{ inputs.ref }}" ]; then
+            echo "ref=" >> $GITHUB_OUTPUT
+          else
+            echo "ref=${{ inputs.ref }}" >> $GITHUB_OUTPUT
+          fi
+
+          if [ -z "${{ inputs.job_filter }}" ]; then
+            echo "job_filter=all" >> $GITHUB_OUTPUT
+          else
+            echo "job_filter=${{ inputs.job_filter }}" >> $GITHUB_OUTPUT
+          fi
+
+          if [ -z "${{ inputs.image_a3 }}" ]; then
+            echo "image_a3=swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11" >> $GITHUB_OUTPUT
+          else
+            echo "image_a3=${{ inputs.image_a3 }}" >> $GITHUB_OUTPUT
+          fi
+
+          if [ -z "${{ inputs.skip_install_flag }}" ]; then
+            echo "skip_install_flag=false" >> $GITHUB_OUTPUT
+          else
+            echo "skip_install_flag=${{ inputs.skip_install_flag }}" >> $GITHUB_OUTPUT
+          fi
+
+  nighly-test-npu:
+    needs: [set-image-config]
+    name: nightly-test-npu
+    if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
+    uses: ./.github/workflows/nightly-test-npu.yml
+    with:
+      ref:  ${{ needs.set-image-config.outputs.ref }}
+      job_filter: ${{ needs.set-image-config.outputs.job_filter }}
+      image_a3: ${{ needs.set-image-config.outputs.image_a3 }}
+      skip_install_flag: ${{ needs.set-image-config.outputs.skip_install_flag }}
+    secrets: inherit
+
+  full-1-npu-a3:
+    needs: [set-image-config]
+    if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
+    runs-on: linux-aarch64-a3-2
+    container:
+      image: ${{ needs.set-image-config.outputs.image_a3 }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ needs.set-image-config.outputs.ref || github.ref }}
+
+      - name: Install dependencies
+        env:
+          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
+          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
+        run: |
+          # speed up by using infra cache services
+          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
+          sed -Ei "s@(ports|archive).ubuntu.com@${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}"
+
+          if [ ${{ needs.set-image-config.outputs.skip_install_flag }} != "true" ];then
+            bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          fi
+
+          # copy required file from our daily cache
+          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
+
+      - name: Print Log Information
+        run: |
+          bash scripts/ci/npu/npu_log_print.sh
+
+      - name: Run test
+        timeout-minutes: 240
+        env:
+          SGLANG_USE_MODELSCOPE: true
+          SGLANG_IS_IN_CI: true
+          HF_ENDPOINT: https://hf-mirror.com
+          TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
+          PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
+          STREAMS_PER_DEVICE: 32
+        run: |
+          pip install sglang_router
+          hf download lmms-lab/MMMU --repo-type dataset
+          pip install sentence_transformers torchaudio==2.8.0
+          pip install protobuf==6.31.1 zss pre-commit wandb>=0.16.0 tenacity==8.3.0 loguru openpyxl latex2sympy2 zstandard transformers-stream-generator tqdm-multiprocess pycocoevalcap
+          pip install yt-dlp sentencepiece==0.1.99 nltk av ftfy sqlitedict==2.1.0 sacrebleu>=1.5.0 pytablewriter black==24.1.0 isort==5.13.2 peft>=0.2.0 accelerate>=0.29.1
+          pip install jsonlines httpx==0.25.0 evaluate>=0.4.0 datasets==2.16.1 numexpr xgrammar==0.1.25 numpy==1.26.4 dotenv
+          git clone --branch v0.3.3 --depth 1 https://github.com/EvolvingLMMs-Lab/lmms-eval.git
+          cd ./lmms-eval
+          nohup pip install . > lmmslog.txt 2>&1 &
+          sleep 120
+          export PYTHONPATH=$PYTHONPATH:$(pwd)
+          cd ../
+          cd test
+          python3 run_suite.py --hw npu --suite full-1-npu-a3 --nightly --continue-on-error --timeout-per-file 3600
+
+  full-2-npu-a3:
+    needs: [set-image-config]
+    if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
+    runs-on: linux-aarch64-a3-2
+    container:
+      image: ${{ needs.set-image-config.outputs.image_a3 }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ needs.set-image-config.outputs.ref || github.ref }}
+
+      - name: Install dependencies
+        env:
+          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
+          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
+        run: |
+          # speed up by using infra cache services
+          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
+          sed -Ei "s@(ports|archive).ubuntu.com@${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}"
+
+          if [ ${{ needs.set-image-config.outputs.skip_install_flag }} != "true" ];then
+            bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          fi
+
+          # copy required file from our daily cache
+          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
+
+      - name: Print Log Information
+        run: |
+          bash scripts/ci/npu/npu_log_print.sh
+
+      - name: Run test
+        timeout-minutes: 240
+        env:
+          SGLANG_USE_MODELSCOPE: true
+          SGLANG_IS_IN_CI: true
+          HF_ENDPOINT: https://hf-mirror.com
+          TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
+          PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
+          STREAMS_PER_DEVICE: 32
+        run: |
+          pip install sglang_router
+          hf download lmms-lab/MMMU --repo-type dataset
+          pip install sentence_transformers torchaudio==2.8.0
+          pip install protobuf==6.31.1 zss pre-commit wandb>=0.16.0 tenacity==8.3.0 loguru openpyxl latex2sympy2 zstandard transformers-stream-generator tqdm-multiprocess pycocoevalcap
+          pip install yt-dlp sentencepiece==0.1.99 nltk av ftfy sqlitedict==2.1.0 sacrebleu>=1.5.0 pytablewriter black==24.1.0 isort==5.13.2 peft>=0.2.0 accelerate>=0.29.1
+          pip install jsonlines httpx==0.25.0 evaluate>=0.4.0 datasets==2.16.1 numexpr xgrammar==0.1.25 numpy==1.26.4 dotenv
+          git clone --branch v0.3.3 --depth 1 https://github.com/EvolvingLMMs-Lab/lmms-eval.git
+          cd ./lmms-eval
+          nohup pip install . > lmmslog.txt 2>&1 &
+          sleep 120
+          export PYTHONPATH=$PYTHONPATH:$(pwd)
+          cd ../
+          cd test
+          python3 run_suite.py --hw npu --suite full-2-npu-a3 --nightly --continue-on-error --timeout-per-file 3600
+
+  full-4-npu-a3:
+    needs: [set-image-config]
+    if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
+    runs-on: linux-aarch64-a3-4
+    container:
+      image: ${{ needs.set-image-config.outputs.image_a3 }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ needs.set-image-config.outputs.ref || github.ref }}
+
+      - name: Install dependencies
+        env:
+          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
+          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
+        run: |
+          # speed up by using infra cache services
+          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
+          sed -Ei "s@(ports|archive).ubuntu.com@${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}"
+
+          if [ ${{ needs.set-image-config.outputs.skip_install_flag }} != "true" ];then
+            bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          fi
+
+          # copy required file from our daily cache
+          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
+
+      - name: Print Log Information
+        run: |
+          bash scripts/ci/npu/npu_log_print.sh
+
+      - name: Run test
+        timeout-minutes: 240
+        env:
+          SGLANG_USE_MODELSCOPE: true
+          SGLANG_IS_IN_CI: true
+          HF_ENDPOINT: https://hf-mirror.com
+          TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
+          PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
+          STREAMS_PER_DEVICE: 32
+        run: |
+          pip install sglang_router
+          hf download lmms-lab/MMMU --repo-type dataset
+          pip install sentence_transformers torchaudio==2.8.0
+          pip install protobuf==6.31.1 zss pre-commit wandb>=0.16.0 tenacity==8.3.0 loguru openpyxl latex2sympy2 zstandard transformers-stream-generator tqdm-multiprocess pycocoevalcap
+          pip install yt-dlp sentencepiece==0.1.99 nltk av ftfy sqlitedict==2.1.0 sacrebleu>=1.5.0 pytablewriter black==24.1.0 isort==5.13.2 peft>=0.2.0 accelerate>=0.29.1
+          pip install jsonlines httpx==0.25.0 evaluate>=0.4.0 datasets==2.16.1 numexpr xgrammar==0.1.25 numpy==1.26.4 dotenv
+          git clone --branch v0.3.3 --depth 1 https://github.com/EvolvingLMMs-Lab/lmms-eval.git
+          cd ./lmms-eval
+          nohup pip install . > lmmslog.txt 2>&1 &
+          sleep 120
+          export PYTHONPATH=$PYTHONPATH:$(pwd)
+          cd ../
+          cd test
+          python3 run_suite.py --hw npu --suite full-4-npu-a3 --nightly --continue-on-error --timeout-per-file 3600
+
+  full-16-npu-a3:
+    needs: [set-image-config]
+    if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
+    runs-on: linux-aarch64-a3-16
+    container:
+      image: ${{ needs.set-image-config.outputs.image_a3 }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ needs.set-image-config.outputs.ref || github.ref }}
+
+      - name: Install dependencies
+        env:
+          TORCH_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/whl/cpu"
+          PYPI_CACHE_URL: "http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
+        run: |
+          # speed up by using infra cache services
+          CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
+          sed -Ei "s@(ports|archive).ubuntu.com@${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}"
+
+          if [ ${{ needs.set-image-config.outputs.skip_install_flag }} != "true" ];then
+            bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          fi
+
+          # copy required file from our daily cache
+          cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
+
+      - name: Print Log Information
+        run: |
+          bash scripts/ci/npu/npu_log_print.sh
+
+      - name: Run test
+        timeout-minutes: 240
+        env:
+          SGLANG_USE_MODELSCOPE: true
+          SGLANG_IS_IN_CI: true
+          HF_ENDPOINT: https://hf-mirror.com
+          TORCH_EXTENSIONS_DIR: /tmp/torch_extensions
+          PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
+          STREAMS_PER_DEVICE: 32
+        run: |
+          pip install sglang_router
+          hf download lmms-lab/MMMU --repo-type dataset
+          pip install sentence_transformers torchaudio==2.8.0
+          pip install protobuf==6.31.1 zss pre-commit wandb>=0.16.0 tenacity==8.3.0 loguru openpyxl latex2sympy2 zstandard transformers-stream-generator tqdm-multiprocess pycocoevalcap
+          pip install yt-dlp sentencepiece==0.1.99 nltk av ftfy sqlitedict==2.1.0 sacrebleu>=1.5.0 pytablewriter black==24.1.0 isort==5.13.2 peft>=0.2.0 accelerate>=0.29.1
+          pip install jsonlines httpx==0.25.0 evaluate>=0.4.0 datasets==2.16.1 numexpr xgrammar==0.1.25 numpy==1.26.4 dotenv
+          git clone --branch v0.3.3 --depth 1 https://github.com/EvolvingLMMs-Lab/lmms-eval.git
+          cd ./lmms-eval
+          nohup pip install . > lmmslog.txt 2>&1 &
+          sleep 120
+          export PYTHONPATH=$PYTHONPATH:$(pwd)
+          cd ../
+          cd test
+          python3 run_suite.py --hw npu --suite full-16-npu-a3 --nightly --continue-on-error --timeout-per-file 3600
+
+  check-all-jobs:
+    if: github.repository == 'sgl-project/sglang' && always()
+    needs:
+      - nighly-test-npu
+      - full-1-npu-a3
+      - full-2-npu-a3
+      - full-4-npu-a3
+      - full-16-npu-a3
+    runs-on: ubuntu-latest
+    container:
+      image: docker.m.daocloud.io/ubuntu:22.04
+    steps:
+      - name: Check if any job failed
+        run: |
+          if [[ "${{ contains(needs.*.result, 'failure') }}" == "true" ]]; then
+            echo "One or more nightly test jobs failed"
+            exit 1
+          fi
+          if [[ "${{ contains(needs.*.result, 'cancelled') }}" == "true" ]]; then
+            echo "One or more nightly test jobs were cancelled"
+            exit 1
+          fi
+          echo "All nightly test jobs passed"
diff --git a/.github/workflows/nightly-test-npu.yml b/.github/workflows/nightly-test-npu.yml
index fa19ab1a4..7503d9a05 100644
--- a/.github/workflows/nightly-test-npu.yml
+++ b/.github/workflows/nightly-test-npu.yml
@@ -2,7 +2,7 @@ name: Nightly Test (NPU)
 
 on:
   schedule:
-    - cron: '0 17 * * *'  # Execute at 1:00 a.m. Beijing Time every day
+    - cron: '0 18 * * *'  # Execute at 2:00 a.m. Beijing Time every day
   pull_request:
     branches:
       - main
@@ -21,13 +21,61 @@ on:
         required: false
         type: string
         default: 'all'
+      image_a3:
+        description: 'The a3 running docker image of the test task.'
+        required: false
+        type: string
+        default: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11'
+      skip_install_flag:
+        description: 'Indicates whether to skip the installation of sglang, defaulting to false.'
+        required: false
+        type: string
+        default: 'false'
+
 
 concurrency:
   group: nightly-test-npu-${{ inputs.ref || github.ref }}
   cancel-in-progress: ${{ github.event_name != 'workflow_call' }}
 
 jobs:
+  set-image-config:
+    runs-on: ubuntu-latest
+    outputs:
+      ref: ${{ steps.set-vars.outputs.ref }}
+      job_filter: ${{ steps.set-vars.outputs.job_filter }}
+      image_a3: ${{ steps.set-vars.outputs.image_a3 }}
+      skip_install_flag: ${{ steps.set-vars.outputs.skip_install_flag }}
+    steps:
+      # When triggered by PR, no inputs parameters are used. The latest community code is tested by default.
+      - name: Set image config
+        id: set-vars
+        run: |
+          if [ -z "${{ inputs.ref }}" ]; then
+            echo "ref=" >> $GITHUB_OUTPUT
+          else
+            echo "ref=${{ inputs.ref }}" >> $GITHUB_OUTPUT
+          fi
+
+          if [ -z "${{ inputs.job_filter }}" ]; then
+            echo "job_filter=all" >> $GITHUB_OUTPUT
+          else
+            echo "job_filter=${{ inputs.job_filter }}" >> $GITHUB_OUTPUT
+          fi
+
+          if [ -z "${{ inputs.image_a3 }}" ]; then
+            echo "image_a3=swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11" >> $GITHUB_OUTPUT
+          else
+            echo "image_a3=${{ inputs.image_a3 }}" >> $GITHUB_OUTPUT
+          fi
+
+          if [ -z "${{ inputs.skip_install_flag }}" ]; then
+            echo "skip_install_flag=false" >> $GITHUB_OUTPUT
+          else
+            echo "skip_install_flag=${{ inputs.skip_install_flag }}" >> $GITHUB_OUTPUT
+          fi
+
   nightly-1-npu-a3:
+    needs: [set-image-config]
     if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
     runs-on: linux-aarch64-a3-2
     strategy:
@@ -35,26 +83,33 @@ jobs:
       matrix:
         part: [0, 1]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
+      image: ${{ needs.set-image-config.outputs.image_a3 }}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
         with:
-          ref: ${{ inputs.ref || github.ref }}
+          ref: ${{ needs.set-image-config.outputs.ref || github.ref }}
 
       - 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"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
         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.extra-index-url "https://pypi.tuna.tsinghua.edu.cn/simple"
-          pip config set global.trusted-host "${CACHING_URL} pypi.tuna.tsinghua.edu.cn"
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          pip config set global.trusted-host "${CACHING_URL}"
+
+          if [ ${{ needs.set-image-config.outputs.skip_install_flag }} != "true" ];then
+            bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          fi
+
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          curl -o /tmp/test.jsonl -L https://raw.githubusercontent.com/openai/grade-school-math/master/grade_school_math/data/test.jsonl
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Print Log Information
         run: |
@@ -86,6 +141,7 @@ jobs:
           python3 run_suite.py --hw npu --suite nightly-1-npu-a3 --nightly --continue-on-error --timeout-per-file 3600 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
 
   nightly-2-npu-a3:
+    needs: [set-image-config]
     if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
     runs-on: linux-aarch64-a3-2
     strategy:
@@ -93,26 +149,33 @@ jobs:
       matrix:
         part: [0]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
+      image: ${{ needs.set-image-config.outputs.image_a3 }}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
         with:
-          ref: ${{ inputs.ref || github.ref }}
+          ref: ${{ needs.set-image-config.outputs.ref || github.ref }}
 
       - 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"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
         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.extra-index-url "https://pypi.tuna.tsinghua.edu.cn/simple"
-          pip config set global.trusted-host "${CACHING_URL} pypi.tuna.tsinghua.edu.cn"
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          pip config set global.trusted-host "${CACHING_URL}"
+
+          if [ ${{ needs.set-image-config.outputs.skip_install_flag }} != "true" ];then
+            bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          fi
+
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          curl -o /tmp/test.jsonl -L https://raw.githubusercontent.com/openai/grade-school-math/master/grade_school_math/data/test.jsonl
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Print Log Information
         run: |
@@ -143,6 +206,7 @@ jobs:
           python3 run_suite.py --hw npu --suite nightly-2-npu-a3 --nightly --continue-on-error --timeout-per-file 3600 --auto-partition-id ${{ matrix.part }} --auto-partition-size 1
 
   nightly-4-npu-a3:
+    needs: [set-image-config]
     if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
     runs-on: linux-aarch64-a3-4
     strategy:
@@ -150,25 +214,33 @@ jobs:
       matrix:
         part: [0]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
+      image: ${{ needs.set-image-config.outputs.image_a3 }}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
         with:
-          ref: ${{ inputs.ref || github.ref }}
+          ref: ${{ needs.set-image-config.outputs.ref|| github.ref }}
 
       - 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"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
         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@${CACHING_URL}:8081@g" /etc/apt/sources.list
-          pip config set global.extra-index-url "https://pypi.tuna.tsinghua.edu.cn/simple"
-          pip config set global.trusted-host "${CACHING_URL} pypi.tuna.tsinghua.edu.cn"
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          pip config set global.index-url http://${CACHING_URL}/pypi/simple
+          pip config set global.trusted-host "${CACHING_URL}"
+
+          if [ ${{ needs.set-image-config.outputs.skip_install_flag }} != "true" ];then
+            bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          fi
+
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          curl -o /tmp/test.jsonl -L https://raw.githubusercontent.com/openai/grade-school-math/master/grade_school_math/data/test.jsonl
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Print Log Information
         run: |
@@ -200,6 +272,7 @@ jobs:
           python3 run_suite.py --hw npu --suite nightly-4-npu-a3 --nightly --continue-on-error --timeout-per-file 3600 --auto-partition-id ${{ matrix.part }} --auto-partition-size 1
 
   nightly-8-npu-a3:
+    needs: [set-image-config]
     if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
     runs-on: linux-aarch64-a3-8
     strategy:
@@ -207,26 +280,33 @@ jobs:
       matrix:
         part: [0]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
+      image: ${{ needs.set-image-config.outputs.image_a3 }}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
         with:
-          ref: ${{ inputs.ref || github.ref }}
+          ref: ${{ needs.set-image-config.outputs.ref || github.ref }}
 
       - 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"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
         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.extra-index-url "https://pypi.tuna.tsinghua.edu.cn/simple"
-          pip config set global.trusted-host "${CACHING_URL} pypi.tuna.tsinghua.edu.cn"
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          pip config set global.trusted-host "${CACHING_URL}"
+
+          if [ ${{ needs.set-image-config.outputs.skip_install_flag }} != "true" ];then
+            bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          fi
+
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          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
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Print Log Information
         run: |
@@ -258,6 +338,7 @@ jobs:
           python3 run_suite.py --hw npu --suite nightly-8-npu-a3 --nightly --continue-on-error --timeout-per-file 3600 --auto-partition-id ${{ matrix.part }} --auto-partition-size 1
 
   nightly-16-npu-a3:
+    needs: [set-image-config]
     if: ${{ (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') }}
     runs-on: linux-aarch64-a3-16
     strategy:
@@ -265,26 +346,33 @@ jobs:
       matrix:
         part: [0, 1]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
+      image: ${{ needs.set-image-config.outputs.image_a3 }}
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
         with:
-          ref: ${{ inputs.ref || github.ref }}
+          ref: ${{ needs.set-image-config.outputs.ref || github.ref }}
 
       - 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"
+          GITHUB_PROXY_URL: "https://gh-proxy.test.osinfra.cn/"
         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.extra-index-url "https://pypi.tuna.tsinghua.edu.cn/simple"
-          pip config set global.trusted-host "${CACHING_URL} pypi.tuna.tsinghua.edu.cn"
-          bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          pip config set global.trusted-host "${CACHING_URL}"
+
+          if [ ${{ needs.set-image-config.outputs.skip_install_flag }} != "true" ];then
+            bash scripts/ci/npu/npu_ci_install_dependency.sh a3
+          fi
+
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          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
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Print Log Information
         run: |
diff --git a/.github/workflows/pr-test-npu.yml b/.github/workflows/pr-test-npu.yml
index 0e79060c0..237f18a27 100644
--- a/.github/workflows/pr-test-npu.yml
+++ b/.github/workflows/pr-test-npu.yml
@@ -76,7 +76,7 @@ jobs:
     uses: ./.github/workflows/pr-gate.yml
     secrets: inherit
 
-  per-commit-1-npu-a2:
+  stage-b-test-1-npu-a2:
     needs: [check-changes, pr-gate]
     if: needs.check-changes.outputs.main_package == 'true'
     runs-on: linux-aarch64-a2-1
@@ -111,21 +111,8 @@ jobs:
           bash scripts/ci/npu/npu_ci_install_dependency.sh 910b
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          curl -o /tmp/test.jsonl -L https://raw.githubusercontent.com/openai/grade-school-math/master/grade_school_math/data/test.jsonl
-
-      - name: Run registered test
-        timeout-minutes: 240
-        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 per-commit-1-npu-a2 --continue-on-error --timeout-per-file 3600 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Run test
         timeout-minutes: 60
@@ -137,10 +124,10 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          cd test/srt
-          python3 run_suite.py --suite per-commit-1-npu-a2 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
+          cd test
+          python3 run_suite.py --hw npu --suite stage-b-test-1-npu-a2 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
 
-  per-commit-2-npu-a2:
+  stage-b-test-2-npu-a2:
     needs: [check-changes, pr-gate]
     if: needs.check-changes.outputs.main_package == 'true'
     runs-on: linux-aarch64-a2-2
@@ -175,8 +162,8 @@ jobs:
           bash scripts/ci/npu/npu_ci_install_dependency.sh 910b
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          curl -o /tmp/test.jsonl -L https://raw.githubusercontent.com/openai/grade-school-math/master/grade_school_math/data/test.jsonl
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Run test
         timeout-minutes: 60
@@ -188,10 +175,10 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          cd test/srt
-          python3 run_suite.py --suite per-commit-2-npu-a2 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
+          cd test
+          python3 run_suite.py --hw npu --suite stage-b-test-2-npu-a2 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
 
-  per-commit-4-npu-a3:
+  stage-b-test-4-npu-a3:
     needs: [check-changes, pr-gate]
     if: needs.check-changes.outputs.main_package == 'true'
     runs-on: linux-aarch64-a3-4
@@ -222,8 +209,8 @@ jobs:
           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 download through proxy
-          curl -o /tmp/test.jsonl -L https://raw.githubusercontent.com/openai/grade-school-math/master/grade_school_math/data/test.jsonl
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Run test
         timeout-minutes: 60
@@ -235,10 +222,11 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          cd test/srt
-          python3 run_suite.py --suite per-commit-4-npu-a3 --timeout-per-file 3600
+          cd test
+          python3 run_suite.py --hw npu --suite stage-b-test-4-npu-a3 --timeout-per-file 3600
 
-  per-commit-16-npu-a3:
+
+  stage-b-test-16-npu-a3:
     needs: [check-changes, pr-gate]
     if: needs.check-changes.outputs.main_package == 'true'
     runs-on: linux-aarch64-a3-16
@@ -269,8 +257,8 @@ jobs:
           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 download through proxy
-          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
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Run test
         timeout-minutes: 60
@@ -282,8 +270,8 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          cd test/srt
-          python3 run_suite.py --suite per-commit-16-npu-a3 --timeout-per-file 3600
+          cd test
+          python3 run_suite.py --hw npu --suite stage-b-test-16-npu-a3 --timeout-per-file 3600
 
   multimodal-gen-test-1-npu-a3:
     needs: [check-changes, pr-gate]
@@ -314,8 +302,8 @@ jobs:
           bash scripts/ci/npu/npu_ci_install_dependency.sh a3 diffusion
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          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
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Run test
         timeout-minutes: 60
@@ -360,8 +348,8 @@ jobs:
           bash scripts/ci/npu/npu_ci_install_dependency.sh a3 diffusion
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          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
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Run test
         timeout-minutes: 60
@@ -380,7 +368,7 @@ jobs:
   multimodal-gen-test-8-npu-a3:
     needs: [check-changes, pr-gate]
     if: needs.check-changes.outputs.multimodal_gen == 'true'
-    runs-on: linux-aarch64-a3-16
+    runs-on: linux-aarch64-a3-8
     container:
       image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
     steps:
@@ -406,8 +394,8 @@ jobs:
           bash scripts/ci/npu/npu_ci_install_dependency.sh a3 diffusion
           # copy required file from our daily cache
           cp ~/.cache/modelscope/hub/datasets/otavia/ShareGPT_Vicuna_unfiltered/ShareGPT_V3_unfiltered_cleaned_split.json /tmp
-          # copy download through proxy
-          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
+          # copy gsm8k dataset
+          cp ~/.cache/modelscope/hub/datasets/tmp/test.jsonl /tmp
 
       - name: Run test
         timeout-minutes: 60
@@ -421,3 +409,45 @@ jobs:
         run: |
           cd python
           python3 sglang/multimodal_gen/test/run_suite.py --suite 8-npu
+
+  pr-test-finish:
+    needs:
+      [
+        check-changes,
+
+        stage-b-test-1-npu-a2,
+        stage-b-test-2-npu-a2,
+        stage-b-test-4-npu-a3,
+        stage-b-test-16-npu-a3,
+
+        multimodal-gen-test-1-npu-a3,
+        multimodal-gen-test-2-npu-a3,
+        multimodal-gen-test-8-npu-a3,
+      ]
+    if: always()
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check all dependent job statuses
+        run: |
+          # Convert the 'needs' context to a JSON string
+          json_needs='${{ toJson(needs) }}'
+
+          # Get a list of all job names from the JSON keys
+          job_names=$(echo "$json_needs" | jq -r 'keys_unsorted[]')
+
+          for job in $job_names; do
+            # For each job, extract its result
+            result=$(echo "$json_needs" | jq -r --arg j "$job" '.[$j].result')
+
+            # Print the job name and its result
+            echo "$job: $result"
+
+            # Check for failure or cancellation and exit if found
+            if [[ "$result" == "failure" || "$result" == "cancelled" ]]; then
+              echo "The above jobs failed."
+              exit 1
+            fi
+          done
+          # If the loop completes, all jobs were successful
+          echo "All jobs completed successfully"
+          exit 0
diff --git a/.github/workflows/release-docker-npu-nightly.yml b/.github/workflows/release-docker-npu-nightly.yml
index 1dc729cfd..8866ae2a2 100644
--- a/.github/workflows/release-docker-npu-nightly.yml
+++ b/.github/workflows/release-docker-npu-nightly.yml
@@ -8,7 +8,7 @@ on:
       - 'docker/npu.Dockerfile'
   workflow_dispatch:
   schedule:
-    - cron: "0 0 * * *"
+    - cron: "0 16 * * *" # Execute at 0:00 a.m. Beijing Time every day
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.sha }}
diff --git a/python/pyproject_npu.toml b/python/pyproject_npu.toml
index ddb0844b2..c7f989467 100644
--- a/python/pyproject_npu.toml
+++ b/python/pyproject_npu.toml
@@ -26,6 +26,8 @@ dependencies = [
   "einops",
   "fastapi",
   "gguf",
+  "hf_transfer",
+  "huggingface_hub",
   "interegular",
   "llguidance>=0.7.11,<0.8.0",
   "modelscope",
diff --git a/python/sglang/test/ascend/test_ascend_utils.py b/python/sglang/test/ascend/test_ascend_utils.py
index 43a078287..681299a30 100644
--- a/python/sglang/test/ascend/test_ascend_utils.py
+++ b/python/sglang/test/ascend/test_ascend_utils.py
@@ -117,9 +117,18 @@ QWEN3_1_7B_GPTQ_INT8_WEIGHTS_PATH = os.path.join(
 QWEN3_235B_A22B_W8A8_WEIGHTS_PATH = os.path.join(
     MODEL_WEIGHTS_DIR, "vllm-ascend/Qwen3-235B-A22B-W8A8"
 )
+QWEN3_30B_A3B_GPTQ_2507_INT4_WEIGHTS_PATH = os.path.join(
+    MODEL_WEIGHTS_DIR, "Qwen/Qwen3-30B-A3B-GPTQ-Int4"
+)
+QWEN3_30B_A3B_INSTRUCT_2507_INT4_AUTOROUND_WEIGHTS_PATH = os.path.join(
+    MODEL_WEIGHTS_DIR, "Intel/Qwen3-30B-A3B-Instruct-2507-int4-AutoRound"
+)
 QWEN3_30B_A3B_INSTRUCT_2507_WEIGHTS_PATH = os.path.join(
     MODEL_WEIGHTS_DIR, "Qwen/Qwen3-30B-A3B-Instruct-2507"
 )
+QWEN3_8B_INT4_AUTOROUND_WEIGHTS_PATH = os.path.join(
+    MODEL_WEIGHTS_DIR, "Intel/Qwen3-8B-int4-AutoRound"
+)
 QWEN3_8B_WEIGHTS_PATH = os.path.join(MODEL_WEIGHTS_DIR, "Qwen/Qwen3-8B")
 QWEN3_8B_EAGLE3_WEIGHTS_PATH = os.path.join(MODEL_WEIGHTS_DIR, "Qwen/Qwen3-8B_eagle3")
 QWEN3_32B_WEIGHTS_PATH = os.path.join(MODEL_WEIGHTS_DIR, "Qwen/Qwen3-32B")
@@ -133,15 +142,6 @@ QWEN3_32B_EAGLE3_WEIGHTS_PATH = os.path.join(MODEL_WEIGHTS_DIR, "Qwen/Qwen3-32B-
 QWEN3_32B_W8A8_MINDIE_WEIGHTS_PATH = os.path.join(
     MODEL_WEIGHTS_DIR, "aleoyang/Qwen3-32B-w8a8-MindIE"
 )
-QWEN3_235B_A22B_W8A8_WEIGHTS_PATH = os.path.join(
-    MODEL_WEIGHTS_DIR, "vllm-ascend/Qwen3-235B-A22B-W8A8"
-)
-QWEN3_CODER_480B_A35B_INSTRUCT_W8A8_QUAROT_WEIGHTS_PATH = os.path.join(
-    MODEL_WEIGHTS_DIR, "Qwen3-Coder-480B-A35B-Instruct-w8a8-QuaRot"
-)
-QWEN3_NEXT_80B_A3B_INSTRUCT_WEIGHTS_PATH = os.path.join(
-    MODEL_WEIGHTS_DIR, "Qwen/Qwen3-Next-80B-A3B-Instruct"
-)
 QWQ_32B_W8A8_WEIGHTS_PATH = os.path.join(MODEL_WEIGHTS_DIR, "vllm-ascend/QWQ-32B-W8A8")
 SMOLLM_1_7B_WEIGHTS_PATH = os.path.join(MODEL_WEIGHTS_DIR, "HuggingFaceTB/SmolLM-1.7B")
 STABLELM_2_1_6B_WEIGHTS_PATH = os.path.join(
diff --git a/test/srt/ascend/test_ascend_hicache_mha.py b/test/registered/ascend/basic_function/HiCache/test_npu_hicache_mha.py
similarity index 91%
rename from test/srt/ascend/test_ascend_hicache_mha.py
rename to test/registered/ascend/basic_function/HiCache/test_npu_hicache_mha.py
index 521537e05..829a5fee4 100644
--- a/test/srt/ascend/test_ascend_hicache_mha.py
+++ b/test/registered/ascend/basic_function/HiCache/test_npu_hicache_mha.py
@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -11,6 +12,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-7B-Instruct": {
         "accuracy": 0.85,
diff --git a/test/srt/ascend/test_ascend_hicache_mla.py b/test/registered/ascend/basic_function/HiCache/test_npu_hicache_mla.py
similarity index 91%
rename from test/srt/ascend/test_ascend_hicache_mla.py
rename to test/registered/ascend/basic_function/HiCache/test_npu_hicache_mla.py
index 4bb355d37..140d590dd 100644
--- a/test/srt/ascend/test_ascend_hicache_mla.py
+++ b/test/registered/ascend/basic_function/HiCache/test_npu_hicache_mla.py
@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -11,6 +12,9 @@ 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="nightly-4-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V2-Lite-W8A8": {
         "accuracy": 0.34,
diff --git a/test/srt/ascend/test_ascend_sampling_backend.py b/test/registered/ascend/basic_function/backends/test_npu_sampling_backend.py
similarity index 92%
rename from test/srt/ascend/test_ascend_sampling_backend.py
rename to test/registered/ascend/basic_function/backends/test_npu_sampling_backend.py
index f0eee21a3..7da4595f2 100644
--- a/test/srt/ascend/test_ascend_sampling_backend.py
+++ b/test/registered/ascend/basic_function/backends/test_npu_sampling_backend.py
@@ -4,6 +4,7 @@ from types import SimpleNamespace
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -12,6 +13,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 
 class TestAscendSamplingBackend(CustomTestCase):
     @classmethod
diff --git a/test/srt/ascend/test_llada2_mini_ascend.py b/test/registered/ascend/basic_function/dllm/test_npu_llada2_mini.py
similarity index 92%
rename from test/srt/ascend/test_llada2_mini_ascend.py
rename to test/registered/ascend/basic_function/dllm/test_npu_llada2_mini.py
index a1fdafcc5..a3c3d137c 100644
--- a/test/srt/ascend/test_llada2_mini_ascend.py
+++ b/test/registered/ascend/basic_function/dllm/test_npu_llada2_mini.py
@@ -3,6 +3,7 @@ import unittest
 from types import SimpleNamespace
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.send_one import BenchArgs, send_one_prompt
 from sglang.test.test_utils import (
@@ -14,6 +15,9 @@ 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="nightly-1-npu-a3", nightly=True)
+
 
 class TestLLaDA2Mini(CustomTestCase):
     @classmethod
diff --git a/test/srt/ascend/test_ascend_compile_graph_tp1_bf16.py b/test/registered/ascend/basic_function/optimization_debug/test_npu_compile_graph_tp1_bf16.py
similarity index 92%
rename from test/srt/ascend/test_ascend_compile_graph_tp1_bf16.py
rename to test/registered/ascend/basic_function/optimization_debug/test_npu_compile_graph_tp1_bf16.py
index e3a51499b..2a94826d6 100644
--- a/test/srt/ascend/test_ascend_compile_graph_tp1_bf16.py
+++ b/test/registered/ascend/basic_function/optimization_debug/test_npu_compile_graph_tp1_bf16.py
@@ -4,6 +4,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -12,6 +13,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-7B-Instruct": {
         "accuracy": 0.84,
diff --git a/test/srt/ascend/test_ascend_graph_tp1_bf16.py b/test/registered/ascend/basic_function/optimization_debug/test_npu_graph_tp1_bf16.py
similarity index 91%
rename from test/srt/ascend/test_ascend_graph_tp1_bf16.py
rename to test/registered/ascend/basic_function/optimization_debug/test_npu_graph_tp1_bf16.py
index 4f8d4b4aa..916e3965d 100644
--- a/test/srt/ascend/test_ascend_graph_tp1_bf16.py
+++ b/test/registered/ascend/basic_function/optimization_debug/test_npu_graph_tp1_bf16.py
@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -11,6 +12,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-7B-Instruct": {
         "accuracy": 0.85,
diff --git a/test/srt/ascend/test_ascend_graph_tp2_bf16.py b/test/registered/ascend/basic_function/optimization_debug/test_npu_graph_tp2_bf16.py
similarity index 91%
rename from test/srt/ascend/test_ascend_graph_tp2_bf16.py
rename to test/registered/ascend/basic_function/optimization_debug/test_npu_graph_tp2_bf16.py
index d4bf902a2..37bb7fd22 100644
--- a/test/srt/ascend/test_ascend_graph_tp2_bf16.py
+++ b/test/registered/ascend/basic_function/optimization_debug/test_npu_graph_tp2_bf16.py
@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -11,6 +12,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-7B-Instruct": {
         "accuracy": 0.85,
diff --git a/test/srt/ascend/test_ascend_piecewise_graph_prefill.py b/test/registered/ascend/basic_function/optimization_debug/test_npu_piecewise_graph_prefill.py
similarity index 92%
rename from test/srt/ascend/test_ascend_piecewise_graph_prefill.py
rename to test/registered/ascend/basic_function/optimization_debug/test_npu_piecewise_graph_prefill.py
index 939486582..6db3aeb87 100644
--- a/test/srt/ascend/test_ascend_piecewise_graph_prefill.py
+++ b/test/registered/ascend/basic_function/optimization_debug/test_npu_piecewise_graph_prefill.py
@@ -2,6 +2,7 @@ import unittest
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -12,6 +13,9 @@ from sglang.test.test_utils import (
     run_bench_one_batch,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 MODEL = "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-7B-Instruct"
 GSM8K_EXP_ACCURACY = 0.84
 EXP_PREFILL_LATENCY = 0.045
diff --git a/test/srt/ascend/test_ascend_deepep.py b/test/registered/ascend/basic_function/parallel_strategy/expert_parallelism/test_npu_deepep.py
similarity index 93%
rename from test/srt/ascend/test_ascend_deepep.py
rename to test/registered/ascend/basic_function/parallel_strategy/expert_parallelism/test_npu_deepep.py
index 19330b862..209cdcf95 100644
--- a/test/srt/ascend/test_ascend_deepep.py
+++ b/test/registered/ascend/basic_function/parallel_strategy/expert_parallelism/test_npu_deepep.py
@@ -4,6 +4,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
@@ -11,6 +12,9 @@ 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="nightly-16-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-R1-0528-W8A8": {
         "accuracy": 0.95,
diff --git a/test/srt/ascend/test_ascend_autoround_dense.py b/test/registered/ascend/basic_function/quant/test_npu_autoround_dense.py
similarity index 86%
rename from test/srt/ascend/test_ascend_autoround_dense.py
rename to test/registered/ascend/basic_function/quant/test_npu_autoround_dense.py
index 475311b6c..87575681b 100644
--- a/test/srt/ascend/test_ascend_autoround_dense.py
+++ b/test/registered/ascend/basic_function/quant/test_npu_autoround_dense.py
@@ -4,6 +4,8 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ascend.test_ascend_utils import QWEN3_8B_INT4_AUTOROUND_WEIGHTS_PATH
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -12,10 +14,13 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 logger = logging.getLogger(__name__)
 
 TEST_MODEL_MATRIX = {
-    "/root/.cache/modelscope/hub/models/Intel/Qwen3-8B-int4-AutoRound": {
+    QWEN3_8B_INT4_AUTOROUND_WEIGHTS_PATH: {
         "accuracy": 0.85,
     },
 }
diff --git a/test/srt/ascend/test_ascend_autoround_moe.py b/test/registered/ascend/basic_function/quant/test_npu_autoround_moe.py
similarity index 85%
rename from test/srt/ascend/test_ascend_autoround_moe.py
rename to test/registered/ascend/basic_function/quant/test_npu_autoround_moe.py
index b0b8f6960..1864ec6ee 100644
--- a/test/srt/ascend/test_ascend_autoround_moe.py
+++ b/test/registered/ascend/basic_function/quant/test_npu_autoround_moe.py
@@ -4,6 +4,10 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ascend.test_ascend_utils import (
+    QWEN3_30B_A3B_INSTRUCT_2507_INT4_AUTOROUND_WEIGHTS_PATH,
+)
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -12,10 +16,13 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 logger = logging.getLogger(__name__)
 
 TEST_MODEL_MATRIX = {
-    "/root/.cache/modelscope/hub/models/Intel/Qwen3-30B-A3B-Instruct-2507-int4-AutoRound": {
+    QWEN3_30B_A3B_INSTRUCT_2507_INT4_AUTOROUND_WEIGHTS_PATH: {
         "accuracy": 0.85,
     },
 }
diff --git a/test/srt/ascend/test_ascend_gptq_moe.py b/test/registered/ascend/basic_function/quant/test_npu_gptq_moe.py
similarity index 86%
rename from test/srt/ascend/test_ascend_gptq_moe.py
rename to test/registered/ascend/basic_function/quant/test_npu_gptq_moe.py
index 22b954379..686f5daa1 100644
--- a/test/srt/ascend/test_ascend_gptq_moe.py
+++ b/test/registered/ascend/basic_function/quant/test_npu_gptq_moe.py
@@ -4,6 +4,10 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ascend.test_ascend_utils import (
+    QWEN3_30B_A3B_GPTQ_2507_INT4_WEIGHTS_PATH,
+)
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -12,10 +16,13 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 logger = logging.getLogger(__name__)
 
 TEST_MODEL_MATRIX = {
-    "/root/.cache/modelscope/hub/models/Qwen/Qwen3-30B-A3B-GPTQ-Int4": {
+    QWEN3_30B_A3B_GPTQ_2507_INT4_WEIGHTS_PATH: {
         "accuracy": 0.85,
     },
 }
diff --git a/test/srt/ascend/test_ascend_w4a4_quantization.py b/test/registered/ascend/basic_function/quant/test_npu_w4a4_quantization.py
similarity index 93%
rename from test/srt/ascend/test_ascend_w4a4_quantization.py
rename to test/registered/ascend/basic_function/quant/test_npu_w4a4_quantization.py
index 22d3f0615..e395ec4c8 100644
--- a/test/srt/ascend/test_ascend_w4a4_quantization.py
+++ b/test/registered/ascend/basic_function/quant/test_npu_w4a4_quantization.py
@@ -12,6 +12,7 @@ from urllib.parse import urlparse
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -21,6 +22,9 @@ 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="nightly-4-npu-a3", nightly=True)
+
 if "ASCEND_RT_VISIBLE_DEVICES" not in os.environ:
     os.environ["ASCEND_RT_VISIBLE_DEVICES"] = "0,1,2,3"
 DEFAULT_PORT_FOR_SRT_TEST_RUNNER = (
diff --git a/test/srt/ascend/test_ascend_w8a8_quantization.py b/test/registered/ascend/basic_function/quant/test_npu_w8a8_quantization.py
similarity index 93%
rename from test/srt/ascend/test_ascend_w8a8_quantization.py
rename to test/registered/ascend/basic_function/quant/test_npu_w8a8_quantization.py
index e0b354570..96bea7efb 100644
--- a/test/srt/ascend/test_ascend_w8a8_quantization.py
+++ b/test/registered/ascend/basic_function/quant/test_npu_w8a8_quantization.py
@@ -12,6 +12,7 @@ from urllib.parse import urlparse
 import requests
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -21,6 +22,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 if "ASCEND_RT_VISIBLE_DEVICES" not in os.environ:
     os.environ["ASCEND_RT_VISIBLE_DEVICES"] = "0,1"
 DEFAULT_PORT_FOR_SRT_TEST_RUNNER = (
diff --git a/test/srt/ascend/test_ascend_mla_fia_w8a8int8.py b/test/registered/ascend/basic_function/runtime_opts/test_npu_mla_fia_w8a8int8.py
similarity index 92%
rename from test/srt/ascend/test_ascend_mla_fia_w8a8int8.py
rename to test/registered/ascend/basic_function/runtime_opts/test_npu_mla_fia_w8a8int8.py
index 4001df6f6..0b4902837 100644
--- a/test/srt/ascend/test_ascend_mla_fia_w8a8int8.py
+++ b/test/registered/ascend/basic_function/runtime_opts/test_npu_mla_fia_w8a8int8.py
@@ -4,6 +4,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -12,6 +13,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V2-Lite-W8A8": {
         "accuracy": 0.34,
diff --git a/test/srt/ascend/test_ascend_mla_w8a8int8.py b/test/registered/ascend/basic_function/runtime_opts/test_npu_mla_w8a8int8.py
similarity index 91%
rename from test/srt/ascend/test_ascend_mla_w8a8int8.py
rename to test/registered/ascend/basic_function/runtime_opts/test_npu_mla_w8a8int8.py
index 177af099a..c50bee071 100644
--- a/test/srt/ascend/test_ascend_mla_w8a8int8.py
+++ b/test/registered/ascend/basic_function/runtime_opts/test_npu_mla_w8a8int8.py
@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -11,6 +12,9 @@ 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="nightly-4-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/vllm-ascend/DeepSeek-V2-Lite-W8A8": {
         "accuracy": 0.34,
diff --git a/test/srt/ascend/test_ascend_tp1_bf16.py b/test/registered/ascend/basic_function/runtime_opts/test_npu_tp1_bf16.py
similarity index 91%
rename from test/srt/ascend/test_ascend_tp1_bf16.py
rename to test/registered/ascend/basic_function/runtime_opts/test_npu_tp1_bf16.py
index abc960953..b01510dc7 100644
--- a/test/srt/ascend/test_ascend_tp1_bf16.py
+++ b/test/registered/ascend/basic_function/runtime_opts/test_npu_tp1_bf16.py
@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -11,6 +12,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-1-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-7B-Instruct": {
         "accuracy": 0.84,
diff --git a/test/srt/ascend/test_ascend_tp2_bf16.py b/test/registered/ascend/basic_function/runtime_opts/test_npu_tp2_bf16.py
similarity index 91%
rename from test/srt/ascend/test_ascend_tp2_bf16.py
rename to test/registered/ascend/basic_function/runtime_opts/test_npu_tp2_bf16.py
index e1f736e9b..8f85a16c0 100644
--- a/test/srt/ascend/test_ascend_tp2_bf16.py
+++ b/test/registered/ascend/basic_function/runtime_opts/test_npu_tp2_bf16.py
@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -11,6 +12,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-7B-Instruct": {
         "accuracy": 0.85,
diff --git a/test/srt/ascend/test_ascend_tp2_fia_bf16.py b/test/registered/ascend/basic_function/runtime_opts/test_npu_tp2_fia_bf16.py
similarity index 92%
rename from test/srt/ascend/test_ascend_tp2_fia_bf16.py
rename to test/registered/ascend/basic_function/runtime_opts/test_npu_tp2_fia_bf16.py
index 5f82bb47f..54f3db7d7 100644
--- a/test/srt/ascend/test_ascend_tp2_fia_bf16.py
+++ b/test/registered/ascend/basic_function/runtime_opts/test_npu_tp2_fia_bf16.py
@@ -4,6 +4,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -12,6 +13,9 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+register_npu_ci(est_time=400, suite="stage-b-test-2-npu-a2", nightly=False)
+register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-7B-Instruct": {
         "accuracy": 0.85,
diff --git a/test/srt/ascend/test_ascend_tp4_bf16.py b/test/registered/ascend/basic_function/runtime_opts/test_npu_tp4_bf16.py
similarity index 91%
rename from test/srt/ascend/test_ascend_tp4_bf16.py
rename to test/registered/ascend/basic_function/runtime_opts/test_npu_tp4_bf16.py
index 79efa2445..85873ad7b 100644
--- a/test/srt/ascend/test_ascend_tp4_bf16.py
+++ b/test/registered/ascend/basic_function/runtime_opts/test_npu_tp4_bf16.py
@@ -3,6 +3,7 @@ from types import SimpleNamespace
 from urllib.parse import urlparse
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
@@ -10,6 +11,9 @@ 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="nightly-4-npu-a3", nightly=True)
+
 TEST_MODEL_MATRIX = {
     "Qwen/Qwen3-30B-A3B-Instruct-2507": {
         "accuracy": 0.90,
diff --git a/test/registered/ascend/embedding_models/test_npu_bge_large_en_v1_5.py b/test/registered/ascend/embedding_models/test_npu_bge_large_en_v1_5.py
index 8e2869d70..5da11e7a4 100644
--- a/test/registered/ascend/embedding_models/test_npu_bge_large_en_v1_5.py
+++ b/test/registered/ascend/embedding_models/test_npu_bge_large_en_v1_5.py
@@ -11,7 +11,7 @@ from sglang.test.test_utils import CustomTestCase, get_similarities
 
 register_npu_ci(
     est_time=400,
-    suite="nightly-1-npu-a3",
+    suite="full-1-npu-a3",
     nightly=True,
     disabled="embeddings are not all close",
 )
diff --git a/test/registered/ascend/llm_models/test_npu_afm_4_5b.py b/test/registered/ascend/llm_models/test_npu_afm_4_5b.py
index ce905093f..9f83350f9 100644
--- a/test/registered/ascend/llm_models/test_npu_afm_4_5b.py
+++ b/test/registered/ascend/llm_models/test_npu_afm_4_5b.py
@@ -5,7 +5,7 @@ from sglang.test.ascend.test_ascend_utils import AFM_4_5B_BASE_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="full-1-npu-a3", nightly=True)
 
 
 class TestAFM(GSM8KAscendMixin, CustomTestCase):
diff --git a/test/registered/ascend/llm_models/test_npu_c4ai_command_r_v01.py b/test/registered/ascend/llm_models/test_npu_c4ai_command_r_v01.py
index c4ee782b4..150572c4d 100644
--- a/test/registered/ascend/llm_models/test_npu_c4ai_command_r_v01.py
+++ b/test/registered/ascend/llm_models/test_npu_c4ai_command_r_v01.py
@@ -8,7 +8,7 @@ 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="nightly-2-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="full-2-npu-a3", nightly=False)
 
 
 class TestC4AI(GSM8KAscendMixin, CustomTestCase):
diff --git a/test/registered/ascend/llm_models/test_npu_exaone_3.py b/test/registered/ascend/llm_models/test_npu_exaone_3.py
index 23e72d4cd..ed676dd5b 100644
--- a/test/registered/ascend/llm_models/test_npu_exaone_3.py
+++ b/test/registered/ascend/llm_models/test_npu_exaone_3.py
@@ -5,7 +5,7 @@ from sglang.test.ascend.test_ascend_utils import EXAONE_3_5_7_8B_INSTRUCT_WEIGHT
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="full-1-npu-a3", nightly=False)
 
 
 class TestEXAONE(GSM8KAscendMixin, CustomTestCase):
diff --git a/test/registered/ascend/llm_models/test_npu_granite_3_0_3b_a800m.py b/test/registered/ascend/llm_models/test_npu_granite_3_0_3b_a800m.py
index 00d3b2a6c..9552e3ad9 100644
--- a/test/registered/ascend/llm_models/test_npu_granite_3_0_3b_a800m.py
+++ b/test/registered/ascend/llm_models/test_npu_granite_3_0_3b_a800m.py
@@ -7,7 +7,7 @@ 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="nightly-1-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="full-1-npu-a3", nightly=True)
 
 
 class TestGranite(GSM8KAscendMixin, CustomTestCase):
diff --git a/test/registered/ascend/llm_models/test_npu_granite_3_1_8b.py b/test/registered/ascend/llm_models/test_npu_granite_3_1_8b.py
index ac665572a..1ef751dd0 100644
--- a/test/registered/ascend/llm_models/test_npu_granite_3_1_8b.py
+++ b/test/registered/ascend/llm_models/test_npu_granite_3_1_8b.py
@@ -5,7 +5,7 @@ from sglang.test.ascend.test_ascend_utils import GRANITE_3_1_8B_INSTRUCT_WEIGHTS
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="full-1-npu-a3", nightly=True)
 
 
 class TestGranite(GSM8KAscendMixin, CustomTestCase):
diff --git a/test/registered/ascend/llm_models/test_npu_grok_2.py b/test/registered/ascend/llm_models/test_npu_grok_2.py
index 3eff4c194..0f75ecc6b 100644
--- a/test/registered/ascend/llm_models/test_npu_grok_2.py
+++ b/test/registered/ascend/llm_models/test_npu_grok_2.py
@@ -6,7 +6,7 @@ from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(
     est_time=400,
-    suite="nightly-16-npu-a3",
+    suite="full-16-npu-a3",
     nightly=False,
     disabled="https://github.com/Ascend/sglang/issues/25",
 )
diff --git a/test/registered/ascend/llm_models/test_npu_ling_lite.py b/test/registered/ascend/llm_models/test_npu_ling_lite.py
index 0dc2a7809..2f24064e8 100644
--- a/test/registered/ascend/llm_models/test_npu_ling_lite.py
+++ b/test/registered/ascend/llm_models/test_npu_ling_lite.py
@@ -5,7 +5,7 @@ from sglang.test.ascend.test_ascend_utils import LING_LITE_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="full-2-npu-a3", nightly=True)
 
 
 class TestLingLite(GSM8KAscendMixin, CustomTestCase):
diff --git a/test/registered/ascend/llm_models/test_npu_mimo_7b_rl.py b/test/registered/ascend/llm_models/test_npu_mimo_7b_rl.py
index 2fe9f802b..e4d354b8b 100644
--- a/test/registered/ascend/llm_models/test_npu_mimo_7b_rl.py
+++ b/test/registered/ascend/llm_models/test_npu_mimo_7b_rl.py
@@ -5,7 +5,7 @@ from sglang.test.ascend.test_ascend_utils import MIMO_7B_RL_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="full-1-npu-a3", nightly=True)
 
 
 class TestMiMo7BRL(GSM8KAscendMixin, CustomTestCase):
diff --git a/test/registered/ascend/llm_models/test_npu_persimmon_8b_chat.py b/test/registered/ascend/llm_models/test_npu_persimmon_8b_chat.py
index 9958edc25..4a4a95782 100644
--- a/test/registered/ascend/llm_models/test_npu_persimmon_8b_chat.py
+++ b/test/registered/ascend/llm_models/test_npu_persimmon_8b_chat.py
@@ -8,7 +8,7 @@ from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(
     est_time=400,
-    suite="nightly-1-npu-a3",
+    suite="full-1-npu-a3",
     nightly=False,
 )
 
diff --git a/test/registered/ascend/llm_models/test_npu_smollm_1_7b.py b/test/registered/ascend/llm_models/test_npu_smollm_1_7b.py
index cfe3722f7..dbe1bcc1c 100644
--- a/test/registered/ascend/llm_models/test_npu_smollm_1_7b.py
+++ b/test/registered/ascend/llm_models/test_npu_smollm_1_7b.py
@@ -5,7 +5,7 @@ from sglang.test.ascend.test_ascend_utils import SMOLLM_1_7B_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="full-1-npu-a3", nightly=True)
 
 
 class TestSmolLM(GSM8KAscendMixin, CustomTestCase):
diff --git a/test/registered/ascend/llm_models/test_npu_stablelm_2_1_6b.py b/test/registered/ascend/llm_models/test_npu_stablelm_2_1_6b.py
index 07c71b070..f61f3ac88 100644
--- a/test/registered/ascend/llm_models/test_npu_stablelm_2_1_6b.py
+++ b/test/registered/ascend/llm_models/test_npu_stablelm_2_1_6b.py
@@ -5,7 +5,7 @@ from sglang.test.ascend.test_ascend_utils import STABLELM_2_1_6B_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="full-1-npu-a3", nightly=True)
 
 
 class TestStablelm(GSM8KAscendMixin, CustomTestCase):
diff --git a/test/registered/ascend/reward_models/test_npu_gemma_2_27b_v0_2.py b/test/registered/ascend/reward_models/test_npu_gemma_2_27b_v0_2.py
index 16772b0ff..086a82342 100644
--- a/test/registered/ascend/reward_models/test_npu_gemma_2_27b_v0_2.py
+++ b/test/registered/ascend/reward_models/test_npu_gemma_2_27b_v0_2.py
@@ -10,7 +10,7 @@ from sglang.test.runners import HFRunner, SRTRunner
 from sglang.test.test_utils import CustomTestCase
 
 logger = logging.getLogger(__name__)
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="full-1-npu-a3", nightly=True)
 
 MODELS = [
     (
diff --git a/test/registered/ascend/reward_models/test_npu_internlm2_7b_reward.py b/test/registered/ascend/reward_models/test_npu_internlm2_7b_reward.py
index 7386e4df9..a0877d30f 100644
--- a/test/registered/ascend/reward_models/test_npu_internlm2_7b_reward.py
+++ b/test/registered/ascend/reward_models/test_npu_internlm2_7b_reward.py
@@ -13,8 +13,8 @@ from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(
     est_time=400,
-    suite="nightly-4-npu-a3",
-    nightly=False,
+    suite="full-4-npu-a3",
+    nightly=True,
 )
 
 PROMPT = (
diff --git a/test/registered/ascend/reward_models/test_npu_llama_3_1_8b_v0_2.py b/test/registered/ascend/reward_models/test_npu_llama_3_1_8b_v0_2.py
index 2f23aaf59..c702e8d0b 100644
--- a/test/registered/ascend/reward_models/test_npu_llama_3_1_8b_v0_2.py
+++ b/test/registered/ascend/reward_models/test_npu_llama_3_1_8b_v0_2.py
@@ -10,7 +10,7 @@ from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.runners import HFRunner, SRTRunner
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=False)
+register_npu_ci(est_time=400, suite="full-1-npu-a3", nightly=True)
 
 MODELS = [
     (SKYWORK_REWARD_LLAMA_3_1_8B_V0_2_WEIGHTS_PATH, 1, 4e-2),
diff --git a/test/registered/ascend/vlm_models/test_npu_qwen3_vl_4b_instruct.py b/test/registered/ascend/vlm_models/test_npu_qwen3_vl_4b_instruct.py
index 802b99d93..33f5b447c 100644
--- a/test/registered/ascend/vlm_models/test_npu_qwen3_vl_4b_instruct.py
+++ b/test/registered/ascend/vlm_models/test_npu_qwen3_vl_4b_instruct.py
@@ -4,7 +4,7 @@ from sglang.test.ascend.test_ascend_utils import QWEN3_VL_4B_INSTRUCT_WEIGHTS_PA
 from sglang.test.ascend.vlm_utils import TestVLMModels
 from sglang.test.ci.ci_register import register_npu_ci
 
-register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="full-4-npu-a3", nightly=True)
 
 
 class TestQwen3VL4B(TestVLMModels):
diff --git a/test/run_suite.py b/test/run_suite.py
index 3762cb0ad..1062f7d2f 100644
--- a/test/run_suite.py
+++ b/test/run_suite.py
@@ -105,6 +105,11 @@ NIGHTLY_SUITES = {
         "nightly-4-npu-a3",
         "nightly-8-npu-a3",
         "nightly-16-npu-a3",
+        "full-1-npu-a3",
+        "full-2-npu-a3",
+        "full-4-npu-a3",
+        "full-8-npu-a3",
+        "full-16-npu-a3",
     ],
 }