[CI][nightly]Add GQA C8 nightly tests. (vllm-project#13016)

### What this PR does / why we need it?
dd GQA C8 nightly tests.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?


- vLLM version: v0.25.1
- vLLM main:
vllm-project/vllm@fe784ff

---------

Signed-off-by: pichangping <1337510399@qq.com>

Author: pichangping

.github/workflows/configs/nightly_config.yaml

@@ -172,6 +172,9 @@ a3:
       - name: qwen3-30b-a3b-w8a8
         os: linux-aarch64-nightly-a3-4
         config_file_path: Qwen3-30B-A3B-W8A8.yaml
+      - name: Qwen3-32B-W8A8C8-A3
+        os: linux-aarch64-nightly-a3-4
+        config_file_path: Qwen3-32B-W8A8C8-A3.yaml
       - name: qwen3-32b-int8-prefix-cache
         os: linux-aarch64-nightly-a3-4
         config_file_path: Prefix-Cache-Qwen3-32B-Int8.yaml
@@ -274,6 +277,9 @@ a3-560t:
       - name: qwen3-30b-a3b-w8a8
         os: linux-aarch64-a3-4-cn12-001
         config_file_path: Qwen3-30B-A3B-W8A8.yaml
+      - name: Qwen3-32B-W8A8C8-A3
+        os: linux-aarch64-a3-4-cn12-001
+        config_file_path: Qwen3-32B-W8A8C8-A3.yaml
       - name: qwen3-32b-int8-prefix-cache
         os: linux-aarch64-a3-4-cn12-001
         config_file_path: Prefix-Cache-Qwen3-32B-Int8.yaml

.github/workflows/misc/model_dataset_list.json

@@ -292,7 +292,8 @@
       "Eco-Tech/Kimi-K2.6-w4a8",
       "lightseekorg/kimi-k2.6-eagle3",
       "RedHatAI/GLM-5.2-speculator.dspark",
-      "Eco-Tech/Qwen3-235B-A22B-w8a8-QuaRot"
+      "Eco-Tech/Qwen3-235B-A22B-w8a8-QuaRot",
+      "Eco-Tech/Qwen3-32B-w8a8c8"
     ],
     "datasets": [
       "vllm-ascend/GSM8K-in1024-bs210",

tests/e2e/nightly/single_node/models/configs/Qwen3-32B-W8A8C8-A3.yaml

@@ -0,0 +1,66 @@
+# ==========================================
+# Shared Configurations
+# ==========================================
+
+_envs: &envs
+  TASK_QUEUE_ENABLE: "1"
+  PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
+  HCCL_OP_EXPANSION_MODE: "AIV"
+  VLLM_ASCEND_ENABLE_FLASHCOMM: "1"
+  SERVER_PORT: "DEFAULT_PORT"
+
+_server_cmd: &server_cmd
+  - "--quantization"
+  - "ascend"
+  - "--no-enable-prefix-caching"
+  - "--tensor-parallel-size"
+  - "4"
+  - "--max-num-seqs"
+  - "80"
+  - "--port"
+  - "$SERVER_PORT"
+  - "--max-model-len"
+  - "40960"
+  - "--max-num-batched-tokens"
+  - "40960"
+  - "--block-size"
+  - "128"
+  - "--trust-remote-code"
+  - "--gpu-memory-utilization"
+  - "0.9"
+
+_benchmarks: &benchmarks
+  acc:
+    case_type: accuracy
+    dataset_path: vllm-ascend/gsm8k-lite
+    request_conf: vllm_api_general_chat
+    dataset_conf: gsm8k/gsm8k_gen_0_shot_noncot_chat_prompt
+    max_out_len: 10240
+    batch_size: 32
+    baseline: 96
+    threshold: 10
+
+# ==========================================
+# ACTUAL TEST CASES
+# ==========================================
+
+test_cases:
+  - name: "Qwen3-32B-W8A8C8-aclgraph-a3"
+    model: "Eco-Tech/Qwen3-32B-w8a8c8"
+    envs:
+      <<: *envs
+    server_cmd: *server_cmd
+    server_cmd_extra:
+      - "--compilation-config"
+      - '{"cudagraph_mode":"FULL_DECODE_ONLY","cudagraph_capture_sizes":[1,12,16,20,24,32,48,60,64,68,72,76,80]}'
+    benchmarks:
+      <<: *benchmarks
+
+  - name: "Qwen3-32B-W8A8C8-single-a3"
+    model: "Eco-Tech/Qwen3-32B-w8a8c8"
+    envs:
+      <<: *envs
+    server_cmd: *server_cmd
+    server_cmd_extra:
+      - "--enforce-eager"
+    benchmarks: