[CI] Append test variant name to markdown report header in nightly test (#16166)

python/sglang/test/nightly_bench_utils.py

@@ -87,7 +87,9 @@ Note: To view the traces through perfetto-ui, please:
         return f"| {self.batch_size} | {self.input_len} | {self.latency:.2f} | {self.input_throughput:.2f} | {self.output_throughput:.2f} | {accept_length} | {itl:.2f} | {input_cost:.2f} | {output_cost:.2f} | {profile_link} |\n"
 
 
-def generate_markdown_report(trace_dir, results: List[BenchmarkResult]) -> str:
+def generate_markdown_report(
+    trace_dir, results: List[BenchmarkResult], variant: Optional[str] = None
+) -> str:
     """Generate a markdown report from a list of BenchmarkResult object from a single run."""
     # Build model header with run_name if it's not "default"
     model_header = results[0].model_path
@@ -99,6 +101,9 @@ def generate_markdown_report(trace_dir, results: List[BenchmarkResult]) -> str:
     if gpu_config:
         model_header += f" [{gpu_config}]"
 
+    if variant:
+        model_header += f" ({variant})"
+
     summary = f"### {model_header}\n"
 
     summary += "| batch size | input len | latency (s) | input throughput (tok/s)  | output throughput (tok/s) | acc length | ITL (ms) | input cost ($/1M) | output cost ($/1M) | profile (extend) | profile (decode)|\n"

python/sglang/test/nightly_utils.py

@@ -309,14 +309,16 @@ class NightlyBenchmarkRunner:
             print(f"  Warning: Could not fetch spec accept length: {e}")
         return None
 
-    def add_report(self, results: List[BenchmarkResult]) -> None:
+    def add_report(
+        self, results: List[BenchmarkResult], variant: Optional[str] = None
+    ) -> None:
         """Add benchmark results to the full report.
 
         Args:
             results: List of BenchmarkResult objects to add to report
         """
         if results:
-            report_part = generate_markdown_report(self.profile_dir, results)
+            report_part = generate_markdown_report(self.profile_dir, results, variant)
             self.full_report += report_part + "\n"
 
     def write_final_report(self) -> None:

python/sglang/test/performance_test_runner.py

@@ -57,6 +57,7 @@ def run_performance_test(
 
     print(f"\n{'='*60}")
     print(f"Running PERFORMANCE test for {model.model_path}")
+    print(f"  Variant: {model.variant}")
     print(f"  Batch sizes: {batch_sizes}")
     print(f"  Input lens: {input_lens}")
     print(f"  Output lens: {output_lens}")
@@ -80,7 +81,7 @@ def run_performance_test(
         )
 
         if success and results:
-            perf_runner.add_report(results)
+            perf_runner.add_report(results, variant=model.variant)
             print(f"✓ Performance test succeeded for {model.model_path}")
 
             # Validate speculative decoding accept length if threshold is set

python/sglang/test/test_utils.py

@@ -1767,11 +1767,13 @@ class ModelLaunchSettings:
         tp_size: int = 1,
         extra_args: Optional[List[str]] = None,
         env: Optional[dict] = None,
+        variant: Optional[str] = None,
     ):
         self.model_path = model_path
         self.tp_size = tp_size
         self.extra_args = list(extra_args) if extra_args else []
         self.env = env
+        self.variant = variant
 
         if self.tp_size > 1 and "--tp" not in self.extra_args:
             self.extra_args.extend(["--tp", str(self.tp_size)])

test/manual/nightly/test_deepseek_v31_perf.py

@@ -72,7 +72,7 @@ class TestNightlyDeepseekV31Performance(unittest.TestCase):
                     if not success:
                         failed_variants.append(variant_config["name"])
 
-                    self.runner.add_report(results)
+                    self.runner.add_report(results, variant=variant_config["name"])
         finally:
             self.runner.write_final_report()
 

test/manual/nightly/test_deepseek_v32_perf.py

@@ -113,7 +113,7 @@ class TestNightlyDeepseekV32Performance(unittest.TestCase):
                     if not success:
                         failed_variants.append(variant_config["name"])
 
-                    self.runner.add_report(results)
+                    self.runner.add_report(results, variant=variant_config["name"])
         finally:
             self.runner.write_final_report()
 

test/registered/8-gpu-models/test_deepseek_v31.py

@@ -48,12 +48,14 @@ class TestDeepseekV31Unified(unittest.TestCase):
                 DEEPSEEK_V31_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args,
+                variant="TP8",
             ),
             # Variant: "mtp" - TP=8 + EAGLE speculative decoding
             ModelLaunchSettings(
                 DEEPSEEK_V31_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args + mtp_args,
+                variant="TP8+MTP",
             ),
         ]
 

test/registered/8-gpu-models/test_deepseek_v32.py

@@ -55,24 +55,28 @@ class TestDeepseekV32Unified(unittest.TestCase):
                 DEEPSEEK_V32_MODEL_PATH,
                 tp_size=8,
                 extra_args=BASE_ARGS + DP_ARGS,
+                variant="DP8",
             ),
             # Variant: "dp+mtp" - DP + EAGLE speculative decoding
             ModelLaunchSettings(
                 DEEPSEEK_V32_MODEL_PATH,
                 tp_size=8,
                 extra_args=BASE_ARGS + DP_ARGS + MTP_ARGS,
+                variant="DP8+MTP",
             ),
             # Variant: "tp" - Pure TP=8 only
             ModelLaunchSettings(
                 DEEPSEEK_V32_MODEL_PATH,
                 tp_size=8,
                 extra_args=BASE_ARGS + TP_ARGS,
+                variant="TP8",
             ),
             # Variant: "tp+mtp" - Pure TP=8 + EAGLE speculative decoding
             ModelLaunchSettings(
                 DEEPSEEK_V32_MODEL_PATH,
                 tp_size=8,
                 extra_args=BASE_ARGS + TP_ARGS + MTP_ARGS,
+                variant="TP8+MTP",
             ),
         ]
 

test/registered/8-gpu-models/test_glm_46.py

@@ -34,6 +34,7 @@ class TestGLM46Unified(unittest.TestCase):
                 GLM_4_6_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args,
+                variant="TP8",
             ),
         ]
 

test/registered/8-gpu-models/test_glm_46_fp8.py

@@ -39,11 +39,13 @@ class TestGLM46FP8Unified(unittest.TestCase):
                 GLM_4_6_FP8_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args,
+                variant="TP8",
             ),
             ModelLaunchSettings(
                 GLM_4_6_FP8_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args + mtp_args,
+                variant="TP8+MTP",
             ),
         ]
 

test/registered/8-gpu-models/test_kimi_k2.py

@@ -35,6 +35,7 @@ class TestKimiK2Unified(unittest.TestCase):
                 KIMI_K2_THINKING_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args,
+                variant="TP8",
             ),
         ]
 

test/registered/8-gpu-models/test_minimax_m2.py

@@ -37,6 +37,7 @@ class TestMiniMaxM2Unified(unittest.TestCase):
                 MINIMAX_M2_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args,
+                variant="TP8+EP8",
             ),
         ]
 

test/registered/8-gpu-models/test_mistral_large3.py

@@ -63,12 +63,14 @@ class TestMistralLarge3Unified(unittest.TestCase):
                 MISTRAL_LARGE3_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args,
+                variant="TP8",
             ),
             # Variant: "eagle" - TP=8 + trtllm_mla + EAGLE with draft model
             ModelLaunchSettings(
                 MISTRAL_LARGE3_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args + eagle_args,
+                variant="TP8+MTP",
             ),
         ]
 

test/registered/8-gpu-models/test_qwen3_235b.py

@@ -34,6 +34,7 @@ class TestQwen3235BUnified(unittest.TestCase):
                 QWEN3_235B_MODEL_PATH,
                 tp_size=8,
                 extra_args=base_args,
+                variant="TP8",
             ),
         ]