[v0.18.0][CI] Fix and simplify the CI for Qwen3 32B (#8093)

### What this PR does / why we need it?
This PR fixes and simplifies the CI configuration for Qwen3 32B.

The main changes are:
- Remove the redundant `Qwen3-32B-Int8-A3-Feature-Stack3.yaml` config
and consolidate the CI setup into `Qwen3-32B-Int8.yaml`.
- Improve runtime stability by adding
`PYTORCH_NPU_ALLOC_CONF=expandable_segments:True` and setting
`--max-num-seqs 80`.
- Update the accuracy benchmark from `aime2024` to `gsm8k-lite`, and
adjust the related dataset config, output length, baseline, and
threshold accordingly.

These changes make the Qwen3 32B CI easier to maintain and more stable
in nightly validation.

---------

Signed-off-by: ZYang6263 <zy626375@gmail.com>

tests/e2e/nightly/single_node/models/configs/Qwen3-32B-Int8-A3-Feature-Stack3.yaml

@@ -1,69 +0,0 @@
-# ==========================================
-# ACTUAL TEST CASES
-# ==========================================
-
-test_cases:
-  - name: "Qwen3-32B-W8A8-a3-feature-stack3"
-    model: "vllm-ascend/Qwen3-32B-W8A8"
-    envs:
-      VLLM_USE: "1"
-      TASK_QUEUE_ENABLE: "1"
-      HCCL_OP_EXPANSION_MODE: "AIV"
-      OMP_PROC_BIND: "false"
-      VLLM_ASCEND_ENABLE_TOPK_OPTIMIZE: "1"
-      VLLM_ASCEND_ENABLE_FLASHCOMM: "1"
-      SERVER_PORT: "DEFAULT_PORT"
-    prompts:
-      - "9.11 and 9.8, which is greater?"
-    api_keyword_args:
-      chat_template_kwargs:
-        enable_thinking: true
-    server_cmd:
-      - "--quantization"
-      - "ascend"
-      - "--tensor-parallel-size"
-      - "4"
-      - "--port"
-      - "$SERVER_PORT"
-      - "--trust-remote-code"
-      - "--reasoning-parser"
-      - "qwen3"
-      - "--distributed_executor_backend"
-      - "mp"
-      - "--gpu-memory-utilization"
-      - "0.9"
-      - "--block-size"
-      - "128"
-      - "--max-num-seqs"
-      - "256"
-      - "--enforce-eager"
-      - "--max-model-len"
-      - "35840"
-      - "--max-num-batched-tokens"
-      - "35840"
-      - "--additional-config"
-      - '{"enable_weight_nz_layout":true, "weight_prefetch_config":{"enabled": true}}'
-      - "--compilation-config"
-      - '{"cudagraph_mode":"FULL_DECODE_ONLY", "cudagraph_capture_sizes":[1,8,24,48,60]}'
-    test_content:
-      - "chat_completion"
-    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: 4
-      perf:
-        case_type: performance
-        dataset_path: vllm-ascend/GSM8K-in3500-bs400
-        request_conf: vllm_api_stream_chat
-        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
-        num_prompts: 240
-        max_out_len: 1500
-        batch_size: 60
-        baseline: 1
-        threshold: 0.97

tests/e2e/nightly/single_node/models/configs/Qwen3-32B-Int8.yaml

@@ -4,6 +4,7 @@
 
 _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"
@@ -14,6 +15,8 @@ _server_cmd: &server_cmd
   - "--no-enable-prefix-caching"
   - "--tensor-parallel-size"
   - "4"
+  - "--max-num-seqs"
+  - "80"
   - "--port"
   - "$SERVER_PORT"
   - "--max-model-len"
@@ -23,8 +26,6 @@ _server_cmd: &server_cmd
   - "--block-size"
   - "128"
   - "--trust-remote-code"
-  - "--reasoning-parser"
-  - "qwen3"
   - "--gpu-memory-utilization"
   - "0.9"
   - "--async-scheduling"
@@ -34,23 +35,23 @@ _server_cmd: &server_cmd
 _benchmarks: &benchmarks
   acc:
     case_type: accuracy
-    dataset_path: vllm-ascend/aime2024
+    dataset_path: vllm-ascend/gsm8k-lite
     request_conf: vllm_api_general_chat
-    dataset_conf: aime2024/aime2024_gen_0_shot_chat_prompt
-    max_out_len: 32768
+    dataset_conf: gsm8k/gsm8k_gen_0_shot_noncot_chat_prompt
+    max_out_len: 10240
     batch_size: 32
-    baseline: 83.33
-    threshold: 7
+    baseline: 96
+    threshold: 4
   perf:
-        case_type: performance
-        dataset_path: vllm-ascend/GSM8K-in3500-bs400
-        request_conf: vllm_api_stream_chat
-        dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
-        num_prompts: 304
-        max_out_len: 1500
-        batch_size: 76
-        baseline: 1
-        threshold: 0.97
+    case_type: performance
+    dataset_path: vllm-ascend/GSM8K-in3500-bs400
+    request_conf: vllm_api_stream_chat
+    dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
+    num_prompts: 304
+    max_out_len: 1500
+    batch_size: 76
+    baseline: 1
+    threshold: 0.97
 
 # ==========================================
 # ACTUAL TEST CASES