diff --git a/.github/workflows/download_model.yaml b/.github/workflows/download_model.yaml
new file mode 100644
index 00000000..4cb5b4a1
--- /dev/null
+++ b/.github/workflows/download_model.yaml
@@ -0,0 +1,75 @@
+name: 'model downloader'
+
+on:
+  pull_request:
+    paths:
+      - '.github/workflows/misc/model_list.json'
+      - '.github/workflows/download_model.yaml'
+    types: [labeled, synchronize]
+
+defaults:
+  run:
+    shell: bash -el {0}
+
+concurrency:
+  group: ascend-${{ github.workflow_ref }}
+  cancel-in-progress: true
+
+jobs:
+  download-models:
+    if: contains(github.event.pull_request.labels.*.name, 'model-download')
+    name: Download models from ModelScope
+    runs-on: linux-aarch64-a3-0
+    container:
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/vllm-ascend:nightly-cpu
+
+    steps:
+      - name: Checkout PR branch
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Install dependencies
+        run: |
+          pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
+          pip install modelscope jq
+
+      - name: Extract new models from PR
+        id: diff
+        run: |
+          set -euo pipefail
+
+          JSON_PATH=".github/workflows/misc/model_list.json"
+
+          git fetch origin main
+
+          git show origin/main:$JSON_PATH > /tmp/models_main.json || \
+            echo '{"models":[]}' > /tmp/models_main.json
+
+          cp $JSON_PATH /tmp/models_pr.json
+
+          jq -r '
+            (.models // []) as $pr
+            | input
+            | (.models // []) as $main
+            | ($pr - $main)[]
+          ' /tmp/models_pr.json /tmp/models_main.json > /tmp/new_models.txt
+
+          echo "New models:"
+          cat /tmp/new_models.txt || true
+
+      - name: Download new models (CLI)
+        if: hashFiles('/tmp/new_models.txt') != ''
+        run: |
+          set -euo pipefail
+
+          while read -r model; do
+            [ -z "$model" ] && continue
+            echo "▶ Downloading $model"
+            modelscope download "$model"
+          done < /tmp/new_models.txt
+
+      - name: Summary
+        run: |
+          echo "Downloaded models:"
+          cat /tmp/new_models.txt || echo "No new models"
diff --git a/.github/workflows/misc/model_list.json b/.github/workflows/misc/model_list.json
new file mode 100644
index 00000000..64d07565
--- /dev/null
+++ b/.github/workflows/misc/model_list.json
@@ -0,0 +1,238 @@
+{
+    "models": [
+      "AngelSlim/Qwen3-32B_eagle3",
+      "Anionex/Qwen3-1.7B-W4A8-V1",
+      "ArthurZ/ilama-3.2-1B",
+      "BAAI/bge-base-en-v1.5",
+      "BAAI/bge-large-zh-v1.5",
+      "BAAI/bge-m3",
+      "BAAI/bge-multilingual-gemma2",
+      "BAAI/bge-reranker-large",
+      "BAAI/bge-reranker-v2-m3",
+      "BAAI/bge-small-en-v1.5",
+      "BAAI/kernel_meta",
+      "ByteDance-Seed/BAGEL-7B-MoT",
+      "DeepSeek-ai/DeepSeek-OCR",
+      "DevQuasar/deepseek-ai.DeepSeek-V3.2-BF16",
+      "Eco-Tech/DeepSeek-V3.1-w8a8-mtp-QuaRot",
+      "Eco-Tech/Qwen3-30B-A3B-w8a8",
+      "Howeee/Qwen2.5-1.5B-apeach",
+      "IntervitensInc/pangu-pro-moe-model",
+      "IntervitensInc/pangu-pro-moe-modelt",
+      "JackFram/llama-160m",
+      "JackFram/llama-68m",
+      "Kwai-Keye/Keye-VL-8B-Preview",
+      "LLM-Research/Llama-3.2-11B-Vision",
+      "LLM-Research/Llama-3.2-1B-Instruct",
+      "LLM-Research/Llama-3.2-3B-Instruct",
+      "LLM-Research/Llama-4-Scout-17B-16E-Instruct",
+      "LLM-Research/Llama-4-Scout-17B-16E-Instruct.",
+      "LLM-Research/Meta-Llama-3-8B-Instruct",
+      "LLM-Research/Meta-Llama-3.1-8B-Instruct",
+      "LLM-Research/Molmo-7B-D-0924",
+      "LLM-Research/Phi-4-mini-instruct",
+      "LLM-Research/gemma-2-9b-it",
+      "LLM-Research/gemma-3-4b-it",
+      "LLM-Research/kernel_meta",
+      "OpenBMB/MiniCPM-2B-dpo-bf16",
+      "OpenBMB/MiniCPM-Llama3-V-2_5",
+      "OpenBMB/MiniCPM3-4B",
+      "OpenBMB/MiniCPM4-0.5B",
+      "OpenGVLab/InternVL2-8B",
+      "OpenGVLab/InternVL2_5-8B",
+      "OpenGVLab/InternVL3-78B",
+      "OpenGVLab/InternVL3-8B",
+      "OpenGVLab/InternVL3_5-8B",
+      "OpenGVLab/InternVL3_5-8B-hf",
+      "PaddlePaddle/ERNIE-4.5-21B-A3B-PT",
+      "PaddlePaddle/PaddleOCR-VL",
+      "QuantTrio/Qwen3-VL-235B-A22B-Instruct-AWQ",
+      "Qwen/QwQ-32B",
+      "Qwen/QwQ-32B-AWQ",
+      "Qwen/Qwen",
+      "Qwen/Qwen-Image",
+      "Qwen/Qwen1.5-MoE-A2.7B",
+      "Qwen/Qwen2-1.5B-Instruct",
+      "Qwen/Qwen2-7B",
+      "Qwen/Qwen2-7B-Instruct",
+      "Qwen/Qwen2-7B-W8A8",
+      "Qwen/Qwen2-Audio-7B-Instruct",
+      "Qwen/Qwen2-VL-2B-Instruct",
+      "Qwen/Qwen2-VL-7B",
+      "Qwen/Qwen2-VL-7B-Instruct",
+      "Qwen/Qwen2.5-0.5B-Instruct",
+      "Qwen/Qwen2.5-0.5B-Instruct-AWQ",
+      "Qwen/Qwen2.5-1.5B-Instruct",
+      "Qwen/Qwen2.5-14B-Instruct",
+      "Qwen/Qwen2.5-32B-Instruct",
+      "Qwen/Qwen2.5-7B",
+      "Qwen/Qwen2.5-7B-Instruct",
+      "Qwen/Qwen2.5-7B-Instruct-1M",
+      "Qwen/Qwen2.5-7b-Instruct",
+      "Qwen/Qwen2.5-Math-PRM-7B",
+      "Qwen/Qwen2.5-Omni-3B",
+      "Qwen/Qwen2.5-Omni-7B",
+      "Qwen/Qwen2.5-VL-32B-Instruct",
+      "Qwen/Qwen2.5-VL-3B-Instruct",
+      "Qwen/Qwen2.5-VL-7B-Instruct",
+      "Qwen/Qwen2.7-7B",
+      "Qwen/Qwen3-0.6B",
+      "Qwen/Qwen3-0.6B-Base",
+      "Qwen/Qwen3-235B-A22B",
+      "Qwen/Qwen3-235B-A22B-Instruct-2507",
+      "Qwen/Qwen3-30B-A3B",
+      "Qwen/Qwen3-30B-A3B-Instruct-2507",
+      "Qwen/Qwen3-30B-A3B-W8A8",
+      "Qwen/Qwen3-32B",
+      "Qwen/Qwen3-32B-AWQ",
+      "Qwen/Qwen3-8B",
+      "Qwen/Qwen3-8B-A3B",
+      "Qwen/Qwen3-8B-Base",
+      "Qwen/Qwen3-8B-W8A8",
+      "Qwen/Qwen3-8B-w4a8",
+      "Qwen/Qwen3-8B-w8a8",
+      "Qwen/Qwen3-Base",
+      "Qwen/Qwen3-Coder-30B-A3B-Instruct",
+      "Qwen/Qwen3-Embedding-0.6B",
+      "Qwen/Qwen3-Embedding-8B",
+      "Qwen/Qwen3-Next-80B-A3B-Instruct",
+      "Qwen/Qwen3-Next-A3B-Instruct",
+      "Qwen/Qwen3-Omni-30B-A3B-Instruct",
+      "Qwen/Qwen3-Reranker-0.6B",
+      "Qwen/Qwen3-VL-235B-A22B-Instruct",
+      "Qwen/Qwen3-VL-2B-Instruct",
+      "Qwen/Qwen3-VL-30B-A3B-Instruct",
+      "Qwen/Qwen3-VL-32B-Instruct",
+      "Qwen/Qwen3-VL-8B-Instruct",
+      "RedHatAI/Qwen3-32B-speculator.eagle3",
+      "RedHatAI/Qwen3-8B-speculator.eagle3",
+      "Shanghai_AI_Laboratory/internlm--chat-7b",
+      "Shanghai_AI_Laboratory/internlm-7b",
+      "Shanghai_AI_Laboratory/internlm-7b-chat",
+      "Shanghai_AI_Laboratory/internlm-7bi-chat",
+      "Shanghai_AI_Laboratory/internlm-chat-7b",
+      "Tencent-Hunyuan/HunyuanOCR",
+      "Tengyunw/qwen3_8b_eagle3",
+      "Tongyi-MAI/Z-Image-Turbo",
+      "baichuan-inc/Baichuan2-7B-Chat",
+      "billy800/Qwen3-30B-A3B-Instruct-2507-AWQ",
+      "deepseek-ai/DeepSeek-OCR",
+      "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
+      "deepseek-ai/DeepSeek-V2",
+      "deepseek-ai/DeepSeek-V2-Lite",
+      "deepseek-ai/DeepSeek-V2-Lite-Chat",
+      "deepseek-ai/Deepseek-V2-Lite",
+      "dengcao/ms-marco-MiniLM-L6-v2",
+      "facebook/opt-125m",
+      "google/gemma-2-9b",
+      "google/gemma-3n-E2B-it",
+      "google/siglip2-base-patch16-224",
+      "hmellor/Ilama-3.2-1B",
+      "ibm-research/PowerMoE-3b",
+      "intfloat/multilingual-e5-small",
+      "jason9693/Qwen2.5-1.5B-apeach",
+      "jinaai/jina-embeddings-v3",
+      "jinaai/jina-embeddings-v4",
+      "jinaai/jina-embeddings-v4-vllm-code",
+      "jinaai/jina-embeddings-v4-vllm-retrieval",
+      "kernel_meta/kernel_meta_temp_2116872659434949099",
+      "llava-hf/LLaVA-NeXT-Video-7B-hf",
+      "llava-hf/llava-1.5-7b-hf",
+      "llava-hf/llava-onevision-qwen2-0.5b-ov-hf",
+      "llava-hf/llava-v1.6-mistral-7b-hf",
+      "meta-llama/Llama-3.2-1B-Instruct",
+      "mistralai/Ministral-3-3B-Instruct-2512-BF16",
+      "mistralai/Ministral-3-8B-Instruct-2512-BF16",
+      "mistralai/Mistral-7B-Instruct-v0.1",
+      "mistralai/Mistral-Small-3.1-24B-Instruct-2503",
+      "mlx-community/DeepSeek-V3-3bit-bf16",
+      "moonshotai/..__temp",
+      "moonshotai/Kimi-Linear-48B-A3B-Instruct",
+      "neuralmagic/Qwen2.5-3B-quantized.w8a8",
+      "nv-community/audio-flamingo-3",
+      "nv-community/audio-flamingo-3-hf",
+      "nvidia/audio-flamingo-3-hf",
+      "openbmb/MiniCPM-2B-sft-bf16",
+      "openbmb/MiniCPM-V-2_6",
+      "openbmb/MiniCPM-V-4_5",
+      "opendatalab/MinerU2.5-2509-1.2B",
+      "rhymes-ai/Aria",
+      "sentence-transformers/all-MiniLM-L12-v2",
+      "swift/Qwen3-235B-A22B-Instruct-2507-AWQ",
+      "tclf90/Qwen3-VL-235B-A22B-Instruct-AWQ",
+      "tencent/HunyuanOCR",
+      "unsloth/DeepSeek-V3.1-BF16",
+      "unsloth/Kimi-K2-Thinking-BF16",
+      "unsloth/gpt-oss-20b-BF16",
+      "vllm-ascend/DeepSeek-R1-0528-W8A8",
+      "vllm-ascend/DeepSeek-R1-W8A8",
+      "vllm-ascend/DeepSeek-R1-fa3-pruning",
+      "vllm-ascend/DeepSeek-R1-w4a8-pruning",
+      "vllm-ascend/DeepSeek-V2-Lite",
+      "vllm-ascend/DeepSeek-V2-Lite-W8A8",
+      "vllm-ascend/DeepSeek-V3-Pruning",
+      "vllm-ascend/DeepSeek-V3-W4A8-Pruing",
+      "vllm-ascend/DeepSeek-V3-W8A8",
+      "vllm-ascend/DeepSeek-V3.1",
+      "vllm-ascend/DeepSeek-V3.1-W4A8-puring",
+      "vllm-ascend/DeepSeek-V3.1-W8A8",
+      "vllm-ascend/DeepSeek-V3.2-W8A8",
+      "vllm-ascend/DeepSeek-V3.2-W8A8-Pruning",
+      "vllm-ascend/EAGLE-LLaMA3.1-Instruct-8B",
+      "vllm-ascend/EAGLE3-LLaMA3.1-Instruct-8B",
+      "vllm-ascend/Kimi-K2-Thinking-Pruning",
+      "vllm-ascend/Llama-2-7b-hf",
+      "vllm-ascend/Llama-3.2-3B-Instruct",
+      "vllm-ascend/Meta-Llama-3-8B-Instruct",
+      "vllm-ascend/QwQ-32B-W8A8",
+      "vllm-ascend/QwQ-32B-w8a8",
+      "vllm-ascend/Qwen2-7B-W8A8",
+      "vllm-ascend/Qwen2-VL-7B-W8A8",
+      "vllm-ascend/Qwen2.5-0.5B-Instruct-W8A8",
+      "vllm-ascend/Qwen2.5-0.5B-Instruct-W8A8-new",
+      "vllm-ascend/Qwen2.5-0.5B-Instruct-fa3",
+      "vllm-ascend/Qwen2.5-0.5B-Instruct-w8a8",
+      "vllm-ascend/Qwen2.5-Omni-7B",
+      "vllm-ascend/Qwen3-0.6B",
+      "vllm-ascend/Qwen3-0.6B-Instruct-W8A8",
+      "vllm-ascend/Qwen3-0.6B-W8A16",
+      "vllm-ascend/Qwen3-0.6B-W8A8",
+      "vllm-ascend/Qwen3-1.7B-W4A8-V1",
+      "vllm-ascend/Qwen3-235B-A22B",
+      "vllm-ascend/Qwen3-235B-A22B-W4A8",
+      "vllm-ascend/Qwen3-235B-A22B-W8A8",
+      "vllm-ascend/Qwen3-235B-A22B-w8a8",
+      "vllm-ascend/Qwen3-30B-A3B",
+      "vllm-ascend/Qwen3-30B-A3B-Puring",
+      "vllm-ascend/Qwen3-30B-A3B-W8A8",
+      "vllm-ascend/Qwen3-30B-A3B-W8A8-Pruning",
+      "vllm-ascend/Qwen3-32B-W4A4",
+      "vllm-ascend/Qwen3-32B-W8A8",
+      "vllm-ascend/Qwen3-8B",
+      "vllm-ascend/Qwen3-8B-W4A8",
+      "vllm-ascend/Qwen3-8B-W8A8",
+      "vllm-ascend/Qwen3-Next-80B-A3B-Instruct-W8A8",
+      "vllm-ascend/Qwen3-Next-80B-A3B-Instruct-W8A8-Pruning",
+      "vllm-ascend/Qwen3-Omni-30B-A3B-Thinking",
+      "vllm-ascend/Qwen3-VL-8B-Instruct",
+      "vllm-ascend/TinyLlama-1.1B-Chat-v0.3",
+      "vllm-ascend/benchmark",
+      "vllm-ascend/ilama-3.2-1B",
+      "vllm-ascend/ilama-text2sql-spider",
+      "vllm-ascend/kernel_meta",
+      "vllm-ascend/llama-160m",
+      "vllm-ascend/llama-160m-accelerator",
+      "vllm-ascend/llama-2-7b-sql-lora-test",
+      "vllm-ascend/llama-68m",
+      "vllm-ascend/llama32-3b-text2sql-spider",
+      "vllm-ascend/pangu-pro-moe-pruing",
+      "vllm-ascend/self_cognition_Alice",
+      "vllm-ascend/self_cognition_Bob",
+      "vllm-ascend/tinyllama-colorist-lora",
+      "vllm-ascend/vllm-eagle-llama-68m-random",
+      "wemaster/deepseek_mtp_main_random_bf16",
+      "wemaster/deepseek_mtp_main_random_w8a8_part",
+      "xlangai/OpenCUA-7B"
+    ]
+  }
+  
\ No newline at end of file