[Test][Misc] Update MiniMax-M2.5-w8a8 test configuration (vllm-project#12298)

### What this PR does / why we need it?

This PR updates the test configuration for the `MiniMax-M2.5-w8a8` model
on Ascend NPU.

### Does this PR introduce _any_ user-facing change?

No.

### How was this patch tested?

Tested via nightly single-node E2E tests for MiniMax-M2.5-w8a8.

- vLLM version: v0.24.0
- vLLM main:
vllm-project/vllm@85c09e9

---------

Signed-off-by: chen-commits <1636718796@qq.com>
Signed-off-by: chen <1636718796@qq.com>

Author: chen-commits

.github/workflows/configs/nightly_config.yaml

@@ -32,6 +32,9 @@ a2:
       - name: gemma4-31b-dense
         os: linux-aarch64-a2b3-4
         config_file_path: Gemma4-31B-Dense.yaml
+      - name: MiniMax-M2.5-w8a8-QuaRot-A2
+        os: linux-aarch64-a2b3-8
+        config_file_path: MiniMax-M2.5-w8a8-QuaRot-A2.yaml
   multi_node:
     test_config:
       - name: multi-node-qwen3-235b-dp

tests/e2e/nightly/single_node/models/configs/MiniMax-M2.5-w8a8-QuaRot-A2.yaml

@@ -6,16 +6,14 @@ test_cases:
   - name: "MiniMax-M2.5-w8a8"
     model: "Eco-Tech/MiniMax-M2.5-w8a8-QuaRot"
     envs:
-      HCCL_BUFFSIZE: "512"
+      HCCL_BUFFSIZE: "1200"
       HCCL_OP_EXPANSION_MODE: "AIV"
       PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
       VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
       OMP_NUM_THREADS: "1"
       TASK_QUEUE_ENABLE: "1"
-      VLLM-ASCEND_BALANCE_SCHEDULING: "1"
-      HCCL_INTRA_PCIE_ENABLE: "1"
-      HCCL_INTRA_ROCE_ENABLE: "0"
-      OMP_PROC_BIND: "false"
+      VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
+      VLLM_ASCEND_ENABLE_NZ: "1"
       VLLM_TORCH_PROFILER_WITH_STACK: "0"
       VLLM_TORCH_PROFILER_DIR: "./profile"
       VLLM_USE_MODELSCOPE: "true"
@@ -30,23 +28,19 @@ test_cases:
       - "0.9"
       - "--quantization"
       - "ascend"
-      - "--additional-config"
-      - '{"enable_cpu_binding":true}'
-      - "--model-loader-extra-config"
-      - '{"enable_multithread_load":true,"num_threads":16}'
       - "--speculative_config"
       - '{"method":"eagle3","model":"vllm-ascend/MiniMax-M2.5-eagle-model-0318","num_speculative_tokens":3}'
       - "--enable-expert-parallel"
       - "--enable-chunked-prefill"
-      - "--enable-prefix-caching"
+      - "--no-enable-prefix-caching"
       - "--max-num-seqs"
-      - "100"
+      - "128"
       - "--max-model-len"
       - "196608"
       - "--seed"
       - "1024"
       - "--max-num-batched-tokens"
-      - "6144"
+      - "16384"
       - "--enable-auto-tool-choice"
       - "--tool-call-parser"
       - "minimax_m2"
@@ -56,7 +50,9 @@ test_cases:
       - "--profiler-config"
       - '{"profiler":"torch","torch_profiler_dir":"./profile","torch_profiler_with_stack":false}'
       - "--compilation-config"
-      - '{"cudagraph_mode":"FULL_DECODE_ONLY","cudagraph_capture_sizes":[4,16,40,80,160,256,400]}'
+      - '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
+      - "--additional-config"
+      - '{"enable_cpu_binding": true, "enable_npugraph_ex": true, "enable_static_kernel": true}'
     benchmarks:
       acc:
         case_type: accuracy
@@ -84,5 +80,5 @@ test_cases:
         max_out_len: 1500
         batch_size: 120
         request_rate: 0
-        baseline: 2042
+        baseline: 1860.7
         threshold: 0.97