minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[CI] Split Kimi K2.5 performance batches by config (#34669)

Browse Source
This commit is contained in:
Baizhou Zhang
2026-08-13 03:20:50 -07:00
committed by GitHub
parent a23670ddbf
commit 9c0d4cba3f
1 changed files with 33 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/gb300/test_kimi_k25_nvfp4.py
+33 -18
View File
@@ -38,6 +38,11 @@ DP_EAGLE_ARGS = [
"--speculative-num-draft-tokens=2", "--speculative-num-draft-tokens=2",
] ]
PERFORMANCE_BATCH_SIZES = {
"TP4+EAGLE3": [1, 8],
"TP4+DP4+DPA+EAGLE3": [16],
}
class TestKimiK25Nvfp4(unittest.TestCase): class TestKimiK25Nvfp4(unittest.TestCase):
"""Kimi-K2.5 NVFP4 + EAGLE3 on GB300 (4x GB300 NVL4, tp=4).""" """Kimi-K2.5 NVFP4 + EAGLE3 on GB300 (4x GB300 NVL4, tp=4)."""
@@ -60,25 +65,35 @@ class TestKimiK25Nvfp4(unittest.TestCase):
), ),
] ]
run_combined_tests( failures = []
models=variants, # Pinned to what `ns eval --benchmarks=mmmu-pro:1` sent implicitly --
test_name="Kimi-K2.5-NVFP4", # 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=0.69,
dataset="mmmu_pro_vision", repeat=1,
baseline_accuracy=0.69, 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"Kimi-K2.5-NVFP4 ({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("Kimi-K2.5-NVFP4 failures:\n" + "\n".join(failures))
if __name__ == "__main__": if __name__ == "__main__":