List your company

LLM Pre-training on GPT-2 (val)

3.405LossStoSignSGDv2
3.399363.437433.47553.51357Apr 16, 2026

Evaluation results

MethodLinks
StoSignSGDv2Precision=FP82026.043.40530.12
StoSignSGDPrecision=FP82026.043.41830.51
LionPrecision=FP82026.043.4631.82
SignSGDPrecision=FP82026.043.54634.67