From 1aa6ab41deb56c2889c7739ad91bec37c9941c60 Mon Sep 17 00:00:00 2001 From: Alison Shao <54658187+alisonshao@users.noreply.github.com> Date: Sat, 7 Mar 2026 01:15:34 -0800 Subject: [PATCH] [Nightly] Replace MiniMax-M2 with MiniMax-M2.5 (#20083) Co-authored-by: Alison Shao --- ...test_minimax_m2.py => test_minimax_m25.py} | 26 +++++++++---------- 1 file changed, 12 insertions(+), 14 deletions(-) rename test/registered/8-gpu-models/{test_minimax_m2.py => test_minimax_m25.py} (65%) diff --git a/test/registered/8-gpu-models/test_minimax_m2.py b/test/registered/8-gpu-models/test_minimax_m25.py similarity index 65% rename from test/registered/8-gpu-models/test_minimax_m2.py rename to test/registered/8-gpu-models/test_minimax_m25.py index 58a1371f0..fa8bf70ec 100644 --- a/test/registered/8-gpu-models/test_minimax_m2.py +++ b/test/registered/8-gpu-models/test_minimax_m25.py @@ -9,32 +9,30 @@ from sglang.test.test_utils import ModelLaunchSettings # 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) -MINIMAX_M2_MODEL_PATH = "MiniMaxAI/MiniMax-M2" +MINIMAX_M25_MODEL_PATH = "MiniMaxAI/MiniMax-M2.5" -class TestMiniMaxM2(unittest.TestCase): - """Unified test class for MiniMax-M2 performance and accuracy. +class TestMiniMaxM25(unittest.TestCase): + """Unified test class for MiniMax-M2.5 performance and accuracy. Single variant with TP=8 + EP=8 configuration. - MiniMax-M2 is a 230B MoE model with 10B active params. Runs BOTH: - 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): - """Run performance and accuracy for MiniMax-M2.""" + def test_minimax_m25(self): + """Run performance and accuracy for MiniMax-M2.5.""" base_args = [ - "--tp=8", - "--ep=8", "--trust-remote-code", - "--model-loader-extra-config", - '{"enable_multithread_load": true}', + "--ep=8", + "--mem-fraction-static=0.85", + "--reasoning-parser=minimax-append-think", ] variants = [ ModelLaunchSettings( - MINIMAX_M2_MODEL_PATH, + MINIMAX_M25_MODEL_PATH, tp_size=8, extra_args=base_args, variant="TP8+EP8", @@ -43,10 +41,10 @@ class TestMiniMaxM2(unittest.TestCase): run_combined_tests( models=variants, - test_name="MiniMax-M2", + test_name="MiniMax-M2.5", accuracy_params=AccuracyTestParams(dataset="gsm8k", baseline_accuracy=0.80), performance_params=PerformanceTestParams( - profile_dir="performance_profiles_minimax_m2", + profile_dir="performance_profiles_minimax_m25", ), )