List your company
ResearchFormerly WizWand

Language Understanding on MMLU-Pro (Score)

81.4MMLU-Pro ScoreGLM-4.5-Air
-0.895220.469941.83563.2001Jul 15, 2024Nov 12, 2024Mar 12, 2025Jul 10, 2025Nov 7, 2025Mar 7, 2026Jul 5, 2026
Updated 3mo ago

Evaluation results

MethodLinks
GLM-4.5-AirNumber of Parameters=110B2026.0181.4
Solar OpenNumber of Parameters=102B2026.0180.4
gpt-oss-120bNumber of Parameters=117B, Optimization Level=high2026.0180.4
gpt-oss-120bNumber of Parameters=117B, Optimization Level=medium2026.0178.6
BaselinePrecision=BF162026.0769
Trinity Large Baseshot=5-shot2026.0266.02
AFM-k5984d3s (Ours)Optimization=QAD + Speculative Decoding + SWA2026.0765.9
Uniform scheduleCompression Step=Step 12026.0654.21
Linear scheduleCompression Step=Step 12026.0649.05
Cosine scheduleCompression Step=Step 12026.0646.39
Qwen2-7BParameters=7B2024.0740
Llama-3-8BParameters=8B2024.0735.4
Uniform scheduleCompression Step=Step 22026.0633.94
Gemma-7BParameters=7B2024.0733.7
Polynomial scheduleCompression Step=Step 12026.0633.16
Mistral-7BParameters=7B2024.0730.9
Qwen1.5-7BParameters=7B2024.0729.9
Linear scheduleCompression Step=Step 22026.0624.97
ARPipeline=Qwen3-0.6B2026.0424.7
Exponential scheduleCompression Step=Step 12026.0621.41
Cosine scheduleCompression Step=Step 22026.0621.26
TIDE-CrossPipeline=Cross-Tokenizer2026.0414.52
TIDE-SharedPipeline=Shared-Tokenizer2026.0414.48
TIDE-SharedPipeline=Cross-Tokenizer2026.0414.02
Uniform scheduleCompression Step=Step 32026.0613.95
BD3LMPipeline=Qwen3-0.6B2026.0413.8
Linear scheduleCompression Step=Step 32026.0613.58
CALMPipeline=Cross-Tokenizer2026.0413.47
Polynomial scheduleCompression Step=Step 22026.0613.38
KLPipeline=Shared-Tokenizer2026.0413.19
TIDE-CrossPipeline=Shared-Tokenizer2026.0412.88
Cosine scheduleCompression Step=Step 32026.0612.57
Exponential scheduleCompression Step=Step 22026.0611.83
Polynomial scheduleCompression Step=Step 32026.0611.04
Exponential scheduleCompression Step=Step 32026.0610.52
Cosine scheduleCompression Step=Step 42026.0610.06
Polynomial scheduleCompression Step=Step 42026.069.61
Polynomial scheduleCompression Step=Step 52026.069.23
Linear scheduleCompression Step=Step 42026.069.16
Exponential scheduleCompression Step=Step 42026.068.99
Exponential scheduleCompression Step=Step 52026.068.63
Uniform scheduleCompression Step=Step 42026.068.44
Linear scheduleCompression Step=Step 52026.065.92
Cosine scheduleCompression Step=Step 52026.062.27