minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Nightly] Replace MiniMax-M2 with MiniMax-M2.5 (#20083)

Browse Source 
Co-authored-by: Alison Shao <alisonshao@mac.lan>
This commit is contained in:
Alison Shao
2026-03-07 01:15:34 -08:00
committed by GitHub
co-authored by Alison Shao
parent 43d6a32045
commit 1aa6ab41de
1 changed files with 12 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/8-gpu-models/test_minimax_m2.py → test/registered/8-gpu-models/test_minimax_m25.py
+12 -14
View File
@@ -9,32 +9,30 @@ from sglang.test.test_utils import ModelLaunchSettings
# Runs on both H200 and B200 via nightly-8-gpu-common suite # Runs on both H200 and B200 via nightly-8-gpu-common suite
register_cuda_ci(est_time=1800, suite="nightly-8-gpu-common", nightly=True) register_cuda_ci(est_time=1800, suite="nightly-8-gpu-common", nightly=True)
MINIMAX_M2_MODEL_PATH = "MiniMaxAI/MiniMax-M2" MINIMAX_M25_MODEL_PATH = "MiniMaxAI/MiniMax-M2.5"
class TestMiniMaxM2(unittest.TestCase): class TestMiniMaxM25(unittest.TestCase):
"""Unified test class for MiniMax-M2 performance and accuracy. """Unified test class for MiniMax-M2.5 performance and accuracy.
Single variant with TP=8 + EP=8 configuration. Single variant with TP=8 + EP=8 configuration.
MiniMax-M2 is a 230B MoE model with 10B active params.
Runs BOTH: Runs BOTH:
- Performance test (using NightlyBenchmarkRunner with extra_bench_args) - Performance test (using NightlyBenchmarkRunner with extra_bench_args)
- Accuracy test (using run_eval with mgsm_en) - Accuracy test (using run_eval with gsm8k)
""" """
def test_minimax_m2(self): def test_minimax_m25(self):
"""Run performance and accuracy for MiniMax-M2.""" """Run performance and accuracy for MiniMax-M2.5."""
base_args = [ base_args = [
"--tp=8",
"--ep=8",
"--trust-remote-code", "--trust-remote-code",
"--model-loader-extra-config", "--ep=8",
'{"enable_multithread_load": true}', "--mem-fraction-static=0.85",
"--reasoning-parser=minimax-append-think",
] ]
variants = [ variants = [
ModelLaunchSettings( ModelLaunchSettings(
MINIMAX_M2_MODEL_PATH, MINIMAX_M25_MODEL_PATH,
tp_size=8, tp_size=8,
extra_args=base_args, extra_args=base_args,
variant="TP8+EP8", variant="TP8+EP8",
@@ -43,10 +41,10 @@ class TestMiniMaxM2(unittest.TestCase):
run_combined_tests( run_combined_tests(
models=variants, models=variants,
test_name="MiniMax-M2", test_name="MiniMax-M2.5",
accuracy_params=AccuracyTestParams(dataset="gsm8k", baseline_accuracy=0.80), accuracy_params=AccuracyTestParams(dataset="gsm8k", baseline_accuracy=0.80),
performance_params=PerformanceTestParams( performance_params=PerformanceTestParams(
profile_dir="performance_profiles_minimax_m2", profile_dir="performance_profiles_minimax_m25",
), ),
) )