ResearchBenchmarksTraining Performance on OLMo-1B (24k sequence length)Follow6,161Training Time (ms)AGoQ6,155.86,190.96,2266,261.1May 1, 2026Evaluation ResultsMethodMethodLinksTraining Time (ms)Memory Usage (MB)AGoQSeq. Len.=24kSeq. Len.=24k2026.056,16166,852COATSeq. Len.=24kSeq. Len.=24k2026.056,29194,100