[E2E] Optimize nightly testcase. (#4886)

### What this PR does / why we need it?
Optimize nightly testcase.
Changes:
- tests/e2e/nightly/multi_node/config/models/Qwen3-235B-A3B.yaml: Add
accuracy and performance benchmark
- tests/e2e/models/configs/Qwen3-8B-Base.yaml: Delete
- tests/e2e/models/configs/internlm-7b.yaml: Change to
internlm3-8b-instruct
- tests/e2e/nightly/models/test_deepseek_r1_w8a8_eplb.py: Change to
DeepSeek-R1-0528-W8A8 model

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: menogrey <1299267905@qq.com>

.github/workflows/vllm_ascend_test_nightly_a3.yaml

@@ -53,7 +53,7 @@ jobs:
             config_file_path: DeepSeek-V3.yaml
             size: 2
           - name: multi-node-qwen3-dp
-            config_file_path: Qwen3-235B-A3B.yaml
+            config_file_path: Qwen3-235B-A22B.yaml
             size: 2
           - name: multi-node-dpsk-4node-pd
             config_file_path: DeepSeek-R1-W8A8.yaml

docs/source/developer_guide/contribution/multi_node_test.md

@@ -142,7 +142,7 @@ currently, the multi-node test workflow defined in the [vllm_ascend_test_nightly
                 config_file_path: tests/e2e/nightly/multi_node/config/models/DeepSeek-V3.yaml
                 size: 2
             - name: multi-node-qwen3-dp
-                config_file_path: tests/e2e/nightly/multi_node/config/models/Qwen3-235B-A3B.yaml
+                config_file_path: tests/e2e/nightly/multi_node/config/models/Qwen3-235B-A22B.yaml
                 size: 2
             - name: multi-node-dpsk-4node-pd
                 config_file_path: tests/e2e/nightly/multi_node/config/models/DeepSeek-R1-W8A8.yaml

tests/e2e/models/configs/Qwen3-8B-Base.yaml

@@ -1,14 +0,0 @@
-model_name: "Qwen/Qwen3-8B-Base"
-hardware: "Atlas A2 Series"
-tasks:
-- name: "gsm8k"
-  metrics:
-  - name: "exact_match,strict-match"
-    value: 0.82
-  - name: "exact_match,flexible-extract"
-    value: 0.83
-- name: "ceval-valid"
-  metrics:
-  - name: "acc,none"
-    value: 0.82
-num_fewshot: 5

tests/e2e/models/configs/accuracy.txt

@@ -12,6 +12,6 @@ InternVL3-8B.yaml
 InternVL3_5-8B.yaml
 ERNIE-4.5-21B-A3B-PT.yaml
 gemma-3-4b-it.yaml
-internlm-7b.yaml
+internlm3-8b-instruct.yaml
 Molmo-7B-D-0924.yaml
 llava-1.5-7b-hf.yaml

tests/e2e/models/configs/internlm3-8b-instruct.yaml

@@ -1,4 +1,4 @@
-model_name: "Shanghai_AI_Laboratory/internlm-7b"
+model_name: "Shanghai_AI_Laboratory/internlm3-8b-instruct"
 hardware: "Atlas A2 Series"
 tasks:
 - name: "ceval-valid"

tests/e2e/models/conftest.py

@@ -21,7 +21,7 @@ def pytest_addoption(parser):
     parser.addoption(
         "--config",
         action="store",
-        default="./tests/e2e/models/configs/Qwen3-8B-Base.yaml",
+        default="./tests/e2e/models/configs/Qwen3-8B.yaml",
         help="Path to the model config YAML file",
     )
     parser.addoption(

tests/e2e/nightly/models/test_deepseek_r1_0528_w8a8_eplb.py

@@ -25,7 +25,7 @@ from tests.e2e.conftest import RemoteOpenAIServer
 from tools.aisbench import run_aisbench_cases
 
 MODELS = [
-    "vllm-ascend/DeepSeek-R1-W8A8",
+    "vllm-ascend/DeepSeek-R1-0528-W8A8",
 ]
 
 prompts = [

tests/e2e/nightly/multi_node/config/models/Qwen3-235B-A22B.yaml

@@ -48,6 +48,17 @@ deployment:
         --no-enable-prefix-caching
         --gpu-memory-utilization 0.9
 benchmarks:
+  perf:
+    case_type: performance
+    dataset_path: vllm-ascend/GSM8K-in3500-bs2800
+    request_conf: vllm_api_stream_chat
+    dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
+    num_prompts: 2800
+    max_out_len: 1500
+    batch_size: 700
+    request_rate: 11.2
+    baseline: 1
+    threshold: 0.97
   acc:
     case_type: accuracy
     dataset_path: vllm-ascend/gsm8k