[NPU] Add GitHub test summary and deduplicate test code. Part 2 (#24689)

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Elizaveta Martirosian <elizaveta.martirosian@gmail.com>
Co-authored-by: ronnie_zheng <zl19940307@163.com>

python/sglang/test/ascend/gsm8k_ascend_mixin.py

@@ -29,6 +29,7 @@ class GSM8KAscendMixin(ABC):
     server_cmd = ""
     gsm8k_num_shots = 5
     num_questions = 200
+    gsm8k_parallel = 128
 
     env = {
         **os.environ,
@@ -82,13 +83,14 @@ class GSM8KAscendMixin(ABC):
                 data_path=None,
                 num_questions=self.num_questions,
                 max_new_tokens=512,
-                parallel=128,
+                parallel=self.gsm8k_parallel,
                 host="http://127.0.0.1",
                 port=int(self.base_url.split(":")[-1]),
             )
             metrics = run_eval(args)
             model_metrics["accuracy"] = metrics["accuracy"]
             model_metrics["output_throughput"] = metrics["output_throughput"]
+            model_metrics["latency"] = metrics["latency"]
             self.assertGreaterEqual(
                 metrics["accuracy"],
                 accuracy_threshold,

python/sglang/test/ascend/test_ascend_utils.py

@@ -58,6 +58,9 @@ DEEPSEEK_CODER_V2_LITE_WEIGHTS_PATH = os.path.join(
 DEEPSEEK_CODER_1_3_B_BASE_PATH = os.path.join(
     MODEL_WEIGHTS_DIR, "deepseek-ai/deepseek-coder-1.3b-base"
 )
+ECO_TECH_QWEN3_32B_W4A4_LAOS_WEIGHTS_PATH = os.path.join(
+    MODEL_WEIGHTS_DIR, "Eco-Tech/Qwen3-32B-w4a4-LAOS"
+)
 ERNIE_4_5_21B_A3B_PT_WEIGHTS_PATH = os.path.join(
     MODEL_WEIGHTS_DIR, "baidu/ERNIE-4.5-21B-A3B-PT"
 )
@@ -154,6 +157,9 @@ QWEN3_32B_W8A8_MINDIE_WEIGHTS_PATH = os.path.join(
     MODEL_WEIGHTS_DIR, "aleoyang/Qwen3-32B-w8a8-MindIE"
 )
 QWQ_32B_W8A8_WEIGHTS_PATH = os.path.join(MODEL_WEIGHTS_DIR, "vllm-ascend/QWQ-32B-W8A8")
+REDHATAI_QWEN2_5_0_5B_INSTRUCT_QUANTIZED_W8A8_WEIGHTS_PATH = os.path.join(
+    MODEL_WEIGHTS_DIR, "RedHatAI/Qwen2.5-0.5B-Instruct-quantized.w8a8"
+)
 SMOLLM_1_7B_WEIGHTS_PATH = os.path.join(MODEL_WEIGHTS_DIR, "HuggingFaceTB/SmolLM-1.7B")
 STABLELM_2_1_6B_WEIGHTS_PATH = os.path.join(
     MODEL_WEIGHTS_DIR, "stabilityai/stablelm-2-1_6b"
@@ -565,7 +571,6 @@ def run_bench_serving(
 
 
 HEADER = """
-### Models
 | Model | Server | Client | Output Throughput | Expected Output Throughput | Latency | Expected Latency | Accuracy | Expected Accuracy | Status |
 | ----- | ------ | ------ | -------- | ------------------ | ------- | ---------------- | -------- | --------- | ------ |
 """

python/sglang/test/ascend/test_mmlu.py

@@ -7,6 +7,8 @@ from sglang.test.run_eval import run_eval
 
 class TestMMLU:
 
+    mmlu_num_examples = 128
+
     def test_mmlu(self):
         accuracy_mmlu_threshold = getattr(self, "accuracy_mmlu", 0.00)
 
@@ -23,12 +25,14 @@ class TestMMLU:
                 base_url=self.base_url,
                 model=self.model,
                 eval_name="mmlu",
-                num_examples=128,
+                num_examples=self.mmlu_num_examples,
                 num_threads=32,
             )
             print("Starting mmlu test...")
             metrics = run_eval(args)
             model_metrics["accuracy"] = metrics["score"]
+            model_metrics["latency"] = metrics.get("latency", "-")
+            model_metrics["output_throughput"] = metrics.get("output_throughput", "-")
             self.assertGreater(metrics["score"], accuracy_mmlu_threshold)
         except Exception as e:
             model_metrics["error"] = e