[AMD] [Qwen 3.5 Day 0] Add Qwen 3.5 nightly accuracy tests (#19479)

2026-03-02 19:42:42 -08:00
parent 060720c573
commit 6b8e62f94f
8 changed files with 367 additions and 1 deletions
--- a/.github/workflows/nightly-test-amd-rocm720.yml
+++ b/.github/workflows/nightly-test-amd-rocm720.yml
@@ -43,11 +43,13 @@ on:
          - 'nightly-8-gpu-deepseek-v32-mtp-rocm720'
          - 'nightly-8-gpu-kimi-k25-rocm720'
          - 'nightly-8-gpu-qwen3-235b-rocm720'
+          - 'nightly-8-gpu-qwen35-rocm720'
          - 'nightly-8-gpu-glm5-rocm720'
          - 'nightly-8-gpu-minimax-m25-rocm720'
          # MI35x ROCm 7.2 jobs
          - 'nightly-test-1-gpu-mi35x-rocm720'
          - 'nightly-8-gpu-mi35x-qwen3-235b-mxfp4-rocm720'
+          - 'nightly-8-gpu-mi35x-qwen35-rocm720'
          - 'nightly-accuracy-8-gpu-mi35x-rocm720'
          - 'nightly-8-gpu-mi35x-grok1-int4-rocm720'
          - 'nightly-8-gpu-mi35x-grok2-rocm720'
@@ -551,6 +553,38 @@ jobs:
          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
          exit ${TEST_EXIT_CODE:-0}

+  # 8-GPU Qwen 3.5 (Accuracy) ROCm 7.2
+  nightly-8-gpu-qwen35-rocm720:
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-qwen35-rocm720')
+    runs-on: linux-mi325-gpu-8
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker (ROCm 7.2)
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version rocm720
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: |
+          bash scripts/ci/amd/amd_ci_install_dependency.sh --skip-aiter-build --skip-test-time-deps
+          bash scripts/ci/amd/amd_ci_exec.sh pip install git+https://github.com/huggingface/transformers.git mistral-common "lm-eval[api]" --upgrade
+
+      - name: Accuracy Test ROCm 7.2 (8-GPU Qwen 3.5)
+        timeout-minutes: 120
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-qwen35 --nightly --timeout-per-file 3600 --continue-on-error || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
  # 8-GPU GLM-5 (Accuracy) ROCm 7.2
  nightly-8-gpu-glm5-rocm720:
    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-glm5-rocm720')
@@ -978,6 +1012,39 @@ jobs:
          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
          exit ${TEST_EXIT_CODE:-0}

+  # MI35x 8-GPU Qwen 3.5 (Accuracy) ROCm 7.2
+  nightly-8-gpu-mi35x-qwen35-rocm720:
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-qwen35-rocm720')
+    runs-on: linux-mi35x-gpu-8
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker (ROCm 7.2)
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version rocm720
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: |
+          bash scripts/ci/amd/amd_ci_install_dependency.sh --skip-aiter-build --skip-test-time-deps
+          bash scripts/ci/amd/amd_ci_exec.sh pip install tabulate
+          bash scripts/ci/amd/amd_ci_exec.sh pip install git+https://github.com/huggingface/transformers.git mistral-common "lm-eval[api]" --upgrade
+
+      - name: Accuracy Test MI35x ROCm 7.2 (8-GPU Qwen 3.5)
+        timeout-minutes: 120
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-mi35x-qwen35 --nightly --timeout-per-file 3600 --continue-on-error || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
  nightly-8-gpu-mi35x-glm5-rocm720:
    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-glm5-rocm720')
    runs-on: linux-mi35x-gpu-8
@@ -1098,6 +1165,7 @@ jobs:
      - nightly-8-gpu-deepseek-v32-mtp-rocm720
      - nightly-8-gpu-kimi-k25-rocm720
      - nightly-8-gpu-qwen3-235b-rocm720
+      - nightly-8-gpu-qwen35-rocm720
      - nightly-8-gpu-glm5-rocm720
      - nightly-8-gpu-minimax-m25-rocm720
      # MI35x ROCm 7.2 jobs
@@ -1112,6 +1180,7 @@ jobs:
      - nightly-perf-8-gpu-mi35x-deepseek-v32-mtp-rocm720
      - nightly-8-gpu-mi35x-kimi-k25-rocm720
      - nightly-8-gpu-mi35x-qwen3-235b-mxfp4-rocm720
+      - nightly-8-gpu-mi35x-qwen35-rocm720
      - nightly-8-gpu-mi35x-glm5-rocm720
      - nightly-8-gpu-mi35x-minimax-m25-rocm720
    runs-on: ubuntu-latest
--- a/.github/workflows/nightly-test-amd.yml
+++ b/.github/workflows/nightly-test-amd.yml
@@ -43,11 +43,13 @@ on:
          - 'nightly-8-gpu-deepseek-v32-mtp'
          - 'nightly-8-gpu-kimi-k25'
          - 'nightly-8-gpu-qwen3-235b'
+          - 'nightly-8-gpu-qwen35'
          - 'nightly-8-gpu-glm5'
          - 'nightly-8-gpu-minimax-m25'
          # MI35x jobs
          - 'nightly-test-1-gpu-mi35x'
          - 'nightly-8-gpu-mi35x-qwen3-235b-mxfp4'
+          - 'nightly-8-gpu-mi35x-qwen35'
          - 'nightly-8-gpu-mi35x-kimi-k25'
          - 'nightly-8-gpu-mi35x-glm5'
          - 'nightly-8-gpu-mi35x-minimax-m25'
@@ -555,6 +557,38 @@ jobs:
          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
          exit ${TEST_EXIT_CODE:-0}

+  # 8-GPU Qwen 3.5 (Accuracy)
+  nightly-8-gpu-qwen35:
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-qwen35')
+    runs-on: linux-mi325-gpu-8
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: |
+          bash scripts/ci/amd/amd_ci_install_dependency.sh
+          bash scripts/ci/amd/amd_ci_exec.sh pip install git+https://github.com/huggingface/transformers.git mistral-common "lm-eval[api]" --upgrade
+
+      - name: Accuracy Test (8-GPU Qwen 3.5)
+        timeout-minutes: 120
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-qwen35 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
  nightly-8-gpu-glm5:
    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-glm5')
    runs-on: linux-mi325-8gpu-sglang
@@ -984,6 +1018,39 @@ jobs:
          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
          exit ${TEST_EXIT_CODE:-0}

+  # MI35x 8-GPU Qwen 3.5 (Accuracy)
+  nightly-8-gpu-mi35x-qwen35:
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-qwen35')
+    runs-on: linux-mi35x-gpu-8
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref || github.ref }}
+
+      - name: Setup docker
+        run: |
+          touch github_summary.md
+          bash scripts/ci/amd/amd_ci_start_container.sh
+        env:
+          GITHUB_WORKSPACE: ${{ github.workspace }}
+
+      - name: Install dependencies
+        run: |
+          bash scripts/ci/amd/amd_ci_install_dependency.sh
+          bash scripts/ci/amd/amd_ci_exec.sh pip install tabulate
+          bash scripts/ci/amd/amd_ci_exec.sh pip install git+https://github.com/huggingface/transformers.git mistral-common "lm-eval[api]" --upgrade
+
+      - name: Accuracy Test MI35x (8-GPU Qwen 3.5)
+        timeout-minutes: 120
+        run: |
+          > github_summary.md  # Clear summary file
+          bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
+            -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
+            python3 run_suite.py --hw amd --suite nightly-amd-accuracy-8-gpu-mi35x-qwen35 --nightly --timeout-per-file 3600 || TEST_EXIT_CODE=$?
+          echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
+          exit ${TEST_EXIT_CODE:-0}
+
  nightly-8-gpu-mi35x-glm5:
    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-glm5')
    runs-on: linux-mi35x-gpu-8
@@ -1104,6 +1171,7 @@ jobs:
      - nightly-8-gpu-deepseek-v32-mtp
      - nightly-8-gpu-kimi-k25
      - nightly-8-gpu-qwen3-235b
+      - nightly-8-gpu-qwen35
      - nightly-8-gpu-glm5
      - nightly-8-gpu-minimax-m25
      # MI35x jobs
@@ -1116,6 +1184,7 @@ jobs:
      - nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp
      - nightly-8-gpu-mi35x-kimi-k25
      - nightly-8-gpu-mi35x-qwen3-235b-mxfp4
+      - nightly-8-gpu-mi35x-qwen35
      - nightly-8-gpu-mi35x-glm5
      - nightly-8-gpu-mi35x-minimax-m25
      # MI35x perf jobs excluded from check - perf failures don't block CI
--- a/docs/basic_usage/popular_model_usage.rst
+++ b/docs/basic_usage/popular_model_usage.rst
@@ -13,6 +13,7 @@ For more usage examples and recipes, visit the `SGLang Cookbook <https://cookboo
   gpt_oss.md
   minimax_m2.md
   qwen3.md
+   qwen3_5.md
   qwen3_vl.md
   deepseek_ocr.md
   llama4.md
--- a/docs/basic_usage/qwen3_5.md
+++ b/docs/basic_usage/qwen3_5.md
@@ -0,0 +1,76 @@
+# Qwen 3.5 Usage
+
+Qwen 3.5 is Alibaba's latest generation LLM featuring a hybrid attention architecture, advanced MoE with shared experts, and native multimodal capabilities.
+
+Key architecture features:
+- **Hybrid Attention**: Gated Delta Networks (linear, O(n) complexity) combined with full attention every 4th layer for high associative recall
+- **MoE with Shared Experts**: Top-8 active out of 64 routed experts plus a dedicated shared expert for universal features
+- **Multimodal**: DeepStack Vision Transformer with Conv3d for native image and video understanding
+
+## Launch Qwen 3.5 with SGLang
+
+### Dense Model
+
+To serve `Qwen/Qwen3.5-397B-A17B` on 8 GPUs:
+
+```bash
+python3 -m sglang.launch_server \
+    --model-path Qwen/Qwen3.5-397B-A17B \
+    --tp 8 \
+    --trust-remote-code
+```
+
+### AMD GPU (MI300X / MI325X / MI35X)
+
+On AMD Instinct GPUs, use the `triton` attention backend. Both the full attention layers and the Gated Delta Net (linear attention) layers use Triton-based kernels on ROCm:
+
+```bash
+SGLANG_USE_AITER=1 python3 -m sglang.launch_server \
+    --model-path Qwen/Qwen3.5-397B-A17B \
+    --tp 8 \
+    --attention-backend triton \
+    --trust-remote-code
+```
+
+```{tip}
+Set `SGLANG_USE_AITER=1` to enable AMD's optimized aiter kernels for MoE and GEMM operations.
+```
+
+### Configuration Tips
+
+- `--attention-backend`: Use `triton` on AMD GPUs for Qwen 3.5. The hybrid attention architecture (Gated Delta Networks + full attention) works best with the Triton backend on ROCm. The linear attention (GDN) layers always use Triton kernels internally via the `GDNAttnBackend`.
+- `--watchdog-timeout`: Increase to `1200` or higher for this large model, as weight loading takes significant time.
+- `--model-loader-extra-config '{"enable_multithread_load": true}'`: Enables parallel weight loading for faster startup.
+
+### Reasoning and Tool Calling
+
+Qwen 3.5 supports reasoning and tool calling via the Qwen3 parsers:
+
+```bash
+python3 -m sglang.launch_server \
+    --model-path Qwen/Qwen3.5-397B-A17B \
+    --tp 8 \
+    --trust-remote-code \
+    --reasoning-parser qwen3 \
+    --tool-call-parser qwen3_coder
+```
+
+## Accuracy Evaluation
+
+You can evaluate the model accuracy using `lm-eval`:
+
+```bash
+pip install lm-eval[api]
+
+lm_eval --model local-completions \
+    --model_args '{"base_url": "http://localhost:8000/v1/completions", "model": "Qwen/Qwen3.5-397B-A17B", "num_concurrent": 256, "max_retries": 10, "max_gen_toks": 2048}' \
+    --tasks gsm8k \
+    --batch_size auto \
+    --num_fewshot 5 \
+    --trust_remote_code
+```
+
+## Additional Resources
+
+- [AMD Day 0 Support for Qwen 3.5 on AMD Instinct GPUs](https://www.amd.com/en/developer/resources/technical-articles/2026/day-0-support-for-qwen-3-5-on-amd-instinct-gpus.html)
+- [HuggingFace Model Card](https://huggingface.co/Qwen/Qwen3.5-397B-A17B)
--- a/docs/supported_models/text_generation/generative_models.md
+++ b/docs/supported_models/text_generation/generative_models.md
@@ -29,7 +29,7 @@ in the GitHub search bar.
 | **Kimi K2** (Thinking, Instruct)    | `moonshotai/Kimi-K2-Instruct`                    | Moonshot AI's 1 trillion parameter MoE model (32B active) with 128K–256K context; state-of-the-art agentic intelligence with stable long-horizon agency across 200–300 sequential tool calls. Features MLA attention and native INT4 quantization. [See Reasoning Parser docs](../advanced_features/separate_reasoning.ipynb)|
 | **Kimi Linear** (48B-A3B)           | `moonshotai/Kimi-Linear-48B-A3B-Instruct`        | Moonshot AI's hybrid linear attention model (48B total, 3B active) with 1M token context; features Kimi Delta Attention (KDA) for up to 6× faster decoding and 75% KV cache reduction vs full attention. |
 | **GPT-OSS**       | `openai/gpt-oss-20b`, `openai/gpt-oss-120b`       | OpenAI’s latest GPT-OSS series for complex reasoning, agentic tasks, and versatile developer use cases.|
-| **Qwen** (3, 3MoE, 3Next, 2.5, 2 series)       | `Qwen/Qwen3-0.6B`, `Qwen/Qwen3-30B-A3B` `Qwen/Qwen3-Next-80B-A3B-Instruct `      | Alibaba’s latest Qwen3 series for complex reasoning, language understanding, and generation tasks; Support for MoE variants along with previous generation 2.5, 2, etc. [SGLang provides Qwen3 specific reasoning parser](../advanced_features/separate_reasoning.ipynb)|
+| **Qwen** (3.5, 3, 3MoE, 3Next, 2.5, 2 series)       | `Qwen/Qwen3.5-397B-A17B`, `Qwen/Qwen3-0.6B`, `Qwen/Qwen3-30B-A3B`      | Alibaba’s latest Qwen3 series for complex reasoning, language understanding, and generation tasks; Support for MoE variants along with previous generation 2.5, 2, etc. [SGLang provides Qwen3 specific reasoning parser](../advanced_features/separate_reasoning.ipynb)|
 | **Llama** (2, 3.x, 4 series)        | `meta-llama/Llama-4-Scout-17B-16E-Instruct`       | Meta's open LLM series, spanning 7B to 400B parameters (Llama 2, 3, and new Llama 4) with well-recognized performance. [SGLang provides Llama-4 model-specific optimizations](../basic_usage/llama4.md)  |
 | **Mistral** (Mixtral, NeMo, Small3) | `mistralai/Mistral-7B-Instruct-v0.2`             | Open 7B LLM by Mistral AI with strong performance; extended into MoE (“Mixtral”) and NeMo Megatron variants for larger scale. |
 | **Gemma** (v1, v2, v3)              | `google/gemma-3-1b-it`                            | Google’s family of efficient multilingual models (1B–27B); Gemma 3 offers a 128K context window, and its larger (4B+) variants support vision input. |
--- a/test/lm_eval_configs/Qwen3.5-397B-A17B.yaml
+++ b/test/lm_eval_configs/Qwen3.5-397B-A17B.yaml
@@ -0,0 +1,13 @@
+model_name: "Qwen/Qwen3.5-397B-A17B"
+tasks:
+- name: "gsm8k"
+  metrics:
+  - name: "exact_match,strict-match"
+    value: 0.9704
+  - name: "exact_match,flexible-extract"
+    value: 0.9697
+limit: 1319
+num_concurrent: 256
+num_fewshot: 5
+gen_kwargs: "max_gen_toks=2048"
+rtol: 0.05
--- a/test/registered/amd/accuracy/mi30x/test_qwen35_eval_amd.py
+++ b/test/registered/amd/accuracy/mi30x/test_qwen35_eval_amd.py
@@ -0,0 +1,64 @@
+"""AMD Qwen 3.5 GSM8K lm-eval Evaluation Test (8-GPU)
+
+Tests Qwen/Qwen3.5-397B-A17B (MoE, Hybrid Attention with Gated Delta Networks)
+with lm-eval GSM8K benchmark on MI325/MI300X, matching the AMD Day 0 article.
+
+Registry: nightly-amd-accuracy-8-gpu-qwen35 suite
+"""
+
+import os
+import unittest
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_amd_ci
+from sglang.test.kits.lm_eval_kit import LMEvalMixin
+from sglang.test.test_utils import (
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    popen_launch_server,
+)
+
+register_amd_ci(est_time=3600, suite="nightly-amd-accuracy-8-gpu-qwen35", nightly=True)
+
+QWEN35_MODEL_PATH = "Qwen/Qwen3.5-397B-A17B"
+SERVER_LAUNCH_TIMEOUT = 3600
+TP_SIZE = 8
+
+
+class TestQwen35EvalAMD(LMEvalMixin, CustomTestCase):
+    """Qwen 3.5 GSM8K lm-eval Test for AMD MI325/MI300X."""
+
+    model_config_name = "lm_eval_configs/Qwen3.5-397B-A17B.yaml"
+
+    @classmethod
+    def setUpClass(cls):
+        cls.model = QWEN35_MODEL_PATH
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        other_args = [
+            "--tp",
+            str(TP_SIZE),
+            "--attention-backend",
+            "triton",
+            "--trust-remote-code",
+            "--model-loader-extra-config",
+            '{"enable_multithread_load": true}',
+            "--watchdog-timeout",
+            "1200",
+        ]
+        env = os.environ.copy()
+        env["SGLANG_USE_AITER"] = "1"
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=SERVER_LAUNCH_TIMEOUT,
+            other_args=other_args,
+            env=env,
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+
+if __name__ == "__main__":
+    unittest.main()
--- a/test/registered/amd/accuracy/mi35x/test_qwen35_eval_mi35x.py
+++ b/test/registered/amd/accuracy/mi35x/test_qwen35_eval_mi35x.py
@@ -0,0 +1,74 @@
+"""MI35x Qwen 3.5 GSM8K lm-eval Evaluation Test (8-GPU)
+
+Tests Qwen/Qwen3.5-397B-A17B (MoE, Hybrid Attention with Gated Delta Networks)
+with lm-eval GSM8K benchmark on MI35x, matching the AMD Day 0 article.
+
+Registry: nightly-amd-accuracy-8-gpu-mi35x-qwen35 suite
+"""
+
+import os
+import unittest
+
+import requests
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_amd_ci
+from sglang.test.kits.lm_eval_kit import LMEvalMixin
+from sglang.test.test_utils import (
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    popen_launch_server,
+)
+
+register_amd_ci(
+    est_time=3600, suite="nightly-amd-accuracy-8-gpu-mi35x-qwen35", nightly=True
+)
+
+QWEN35_MODEL_PATH = "Qwen/Qwen3.5-397B-A17B"
+SERVER_LAUNCH_TIMEOUT = 3600
+TP_SIZE = 8
+
+
+class TestQwen35EvalMI35x(LMEvalMixin, CustomTestCase):
+    """Qwen 3.5 GSM8K lm-eval Test for AMD MI35x."""
+
+    model_config_name = "lm_eval_configs/Qwen3.5-397B-A17B.yaml"
+
+    @classmethod
+    def setUpClass(cls):
+        cls.model = QWEN35_MODEL_PATH
+        cls.base_url = DEFAULT_URL_FOR_TEST
+
+    def test_lm_eval(self):
+        """Override to handle server lifecycle within test method (MI35x pattern)."""
+        other_args = [
+            "--tp",
+            str(TP_SIZE),
+            "--attention-backend",
+            "triton",
+            "--trust-remote-code",
+            "--model-loader-extra-config",
+            '{"enable_multithread_load": true}',
+            "--watchdog-timeout",
+            "1200",
+        ]
+        env = os.environ.copy()
+        env["SGLANG_USE_AITER"] = "1"
+
+        process = popen_launch_server(
+            QWEN35_MODEL_PATH,
+            self.base_url,
+            timeout=SERVER_LAUNCH_TIMEOUT,
+            other_args=other_args,
+            env=env,
+        )
+
+        try:
+            requests.get(self.base_url + "/flush_cache")
+            super().test_lm_eval()
+        finally:
+            kill_process_tree(process.pid)
+
+
+if __name__ == "__main__":
+    unittest.main()