[Feature] Xiaomi MiMo-V2.5 day0 support (#23811)

Co-authored-by: 张袁 <zhangyuan36@xiaomi.com>
Co-authored-by: 刘安岐 <liuanqi6@xiaomi.com>
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
Co-authored-by: Shangming Cai <csmthu@gmail.com>
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
This commit is contained in:
Zhonghua Deng
2026-05-01 00:02:26 +08:00
committed by GitHub
co-authored by 张袁 刘安岐 Xinyuan Tong Shangming Cai Xinyuan Tong
parent cf4f462094
commit 651af06a0b
16 changed files with 4369 additions and 87 deletions
@@ -4,6 +4,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.kits.eval_accuracy_kit import GSM8KMixin
from sglang.test.kits.spec_decoding_kit import SpecDecodingMixin
from sglang.test.server_fixtures.default_fixture import DefaultServerBase
from sglang.test.server_fixtures.mmmu_fixture import MMMUServerBase
register_cuda_ci(est_time=330, suite="stage-c-test-8-gpu-h200")
@@ -48,5 +49,42 @@ class TestMiMoV2Flash(GSM8KMixin, SpecDecodingMixin, DefaultServerBase):
num_accepted_drafts_thres = 3.2
MIMO_V2_MODEL = "XiaomiMiMo/MiMo-V2.5"
MIMO_V2_OTHER_ARGS = [
"--tp",
"8",
"--dp",
"2",
"--enable-dp-attention",
"--mm-enable-dp-encoder",
"--attention-backend",
"fa3",
"--mm-attention-backend",
"fa3",
"--reasoning-parser",
"mimo",
]
MIMO_V2_MTP_OTHER_ARGS = MIMO_V2_OTHER_ARGS + [
"--speculative-algorithm",
"EAGLE",
"--speculative-num-steps",
"3",
"--speculative-eagle-topk",
"1",
"--speculative-num-draft-tokens",
"4",
"--enable-multi-layer-eagle",
]
class TestMiMoV2(GSM8KMixin, MMMUServerBase):
gsm8k_accuracy_thres = 0.75
gsm8k_accept_length_thres = 2.5
model = MIMO_V2_MODEL
mem_fraction_static = 0.65
server_api_key = None
other_args = MIMO_V2_MTP_OTHER_ARGS
if __name__ == "__main__":
unittest.main()