From 646fbac7a9e45451f87190e94a663977cd02a389 Mon Sep 17 00:00:00 2001
From: zhangxinyuehfad <59153331+zhangxinyuehfad@users.noreply.github.com>
Date: Tue, 4 Nov 2025 09:23:11 +0800
Subject: [PATCH] [Test] Add accuracy test for qwen3-8b-w8a8 (#3799)

### What this PR does / why we need it?
Add accuracy test for qwen3-8b-w8a8

- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/c9461e05a4ed3557cfbf4b15ded1e26761cc39ca

Signed-off-by: hfadzxy <starmoon_zhang@163.com>
---
 .github/workflows/accuracy_test.yaml         |  2 ++
 tests/e2e/models/configs/Qwen3-8B-W8A8.yaml  | 12 ++++++++++++
 tests/e2e/models/test_lm_eval_correctness.py |  3 ++-
 3 files changed, 16 insertions(+), 1 deletion(-)
 create mode 100644 tests/e2e/models/configs/Qwen3-8B-W8A8.yaml

diff --git a/.github/workflows/accuracy_test.yaml b/.github/workflows/accuracy_test.yaml
index 695cf70e..0822fc59 100644
--- a/.github/workflows/accuracy_test.yaml
+++ b/.github/workflows/accuracy_test.yaml
@@ -59,6 +59,8 @@ jobs:
             model_name: DeepSeek-V2-Lite
           - runner: a2-4
             model_name: Qwen3-Next-80B-A3B-Instruct
+          - runner: a2-1  
+            model_name: Qwen3-8B-W8A8
       fail-fast: false
     # test will be triggered when tag 'accuracy-test' & 'ready-for-test'
     if:  >-
diff --git a/tests/e2e/models/configs/Qwen3-8B-W8A8.yaml b/tests/e2e/models/configs/Qwen3-8B-W8A8.yaml
new file mode 100644
index 00000000..0ab3f756
--- /dev/null
+++ b/tests/e2e/models/configs/Qwen3-8B-W8A8.yaml
@@ -0,0 +1,12 @@
+model_name: "vllm-ascend/Qwen3-8B-W8A8"
+hardware: "Atlas A2 Series"
+tasks:
+- name: "gsm8k"
+  metrics:
+  - name: "exact_match,strict-match"
+    value: 0.80
+  - name: "exact_match,flexible-extract"
+    value: 0.82
+num_fewshot: 5
+enable_thinking: False
+quantization: ascend
diff --git a/tests/e2e/models/test_lm_eval_correctness.py b/tests/e2e/models/test_lm_eval_correctness.py
index b171e59d..a0862b80 100644
--- a/tests/e2e/models/test_lm_eval_correctness.py
+++ b/tests/e2e/models/test_lm_eval_correctness.py
@@ -48,7 +48,8 @@ def build_model_args(eval_config, tp_size):
     }
     for s in [
             "max_images", "gpu_memory_utilization", "enable_expert_parallel",
-            "tensor_parallel_size", "enforce_eager", "enable_thinking"
+            "tensor_parallel_size", "enforce_eager", "enable_thinking",
+            "quantization"
     ]:
         val = eval_config.get(s, None)
         if val is not None: