List your company

Mathematical Reasoning (Fine-tuning) on GSM8K (test)

4.598Test PerplexityBAOC0.5
4.259966.541738.823511.10527May 6, 2026

Evaluation results

MethodLinks
BAOC0.5Architecture=GPT2-small, Time (min)=30.8, StateMem (MB)=2352026.054.598
AdamW16Architecture=GPT2-small, Time (min)=31.5, StateMem (MB)=4752026.054.615
COSMOSArchitecture=GPT2-small, Time (min)=40.3, StateMem (MB)=6602026.054.737
Adam-miniArchitecture=GPT2-small, Time (min)=30.2, StateMem (MB)=4752026.054.995
MuonArchitecture=GPT2-small, Time (min)=35.8, StateMem (MB)=6252026.056.238
Adam-miniArchitecture=T5-base, Time (min)=28.9, StateMem (MB)=7562026.058.898
BAOC0.5Architecture=T5-base, Time (min)=22.8, StateMem (MB)=3572026.059.137
AdamW16Architecture=T5-base, Time (min)=28.2, StateMem (MB)=7562026.059.312
COSMOSArchitecture=T5-base, Time (min)=61.3, StateMem (MB)=8582026.059.921
MuonArchitecture=T5-base, Time (min)=39.2, StateMem (MB)=9452026.0513.049