[CI] Trim Qwen3.5 FP8 GB300 performance batches (#34882)

test/registered/gb300/test_qwen35_fp8.py

@@ -36,6 +36,11 @@ DP_MTP_ARGS = [
     "--speculative-num-draft-tokens=2",
 ]
 
+PERFORMANCE_BATCH_SIZES = {
+    "TP4+MTP": [1, 4],
+    "TP4+DP4+DPA+MTP": [16],
+}
+
 
 class TestQwen35Fp8(unittest.TestCase):
     """Qwen3.5-397B FP8 on GB300 (4x GB300 NVL4, tp=4)."""
@@ -58,25 +63,35 @@ class TestQwen35Fp8(unittest.TestCase):
             ),
         ]
 
-        run_combined_tests(
+        failures = []
-            models=variants,
+        # Pinned to what `ns eval --benchmarks=mmmu-pro:1` sent implicitly --
-            test_name="Qwen3.5-397B-FP8",
+        # its `:1` suffix means temperature 0.7, not greedy -- so the baseline
-            # Pinned to what `ns eval --benchmarks=mmmu-pro:1` sent implicitly --
+        # carries over unchanged. Do not "simplify" these away.
-            # its `:1` suffix means temperature 0.7, not greedy -- so the baseline
+        accuracy_params = AccuracyTestParams(
-            # carries over unchanged. Do not "simplify" these away.
+            dataset="mmmu_pro_vision",
-            accuracy_params=AccuracyTestParams(
+            baseline_accuracy=0.76,
-                dataset="mmmu_pro_vision",
+            repeat=1,
-                baseline_accuracy=0.76,
+            max_tokens=32768,
-                repeat=1,
+            temperature=0.7,
-                max_tokens=32768,
+            seed=0,
-                temperature=0.7,
+            sgl_eval_thinking=False,
-                seed=0,
-                sgl_eval_thinking=False,
-            ),
-            performance_params=PerformanceTestParams(
-                result_dir="performance_results_gb300",
-            ),
         )
+        for variant in variants:
+            try:
+                run_combined_tests(
+                    models=[variant],
+                    test_name=f"Qwen3.5-397B-FP8 ({variant.variant})",
+                    accuracy_params=accuracy_params,
+                    performance_params=PerformanceTestParams(
+                        batch_sizes=PERFORMANCE_BATCH_SIZES[variant.variant],
+                        result_dir="performance_results_gb300",
+                    ),
+                )
+            except AssertionError as e:
+                failures.append(f"{variant.variant}: {e}")
+
+        if failures:
+            raise AssertionError("Qwen3.5-397B-FP8 failures:\n" + "\n".join(failures))
 
 
 if __name__ == "__main__":