List your company

Optimizer Memory Footprint Analysis on OpenWebText GPT-2 125M (train)

0.25Peak Memory FootprintGefen
0.13960.88481.632.3752Jun 11, 2026

Evaluation results

MethodLinks
Gefen2026.060.250.25
Adam4bit2026.060.30.3
Adam8bit2026.060.510.51
Adam-mini2026.061.021.01
SM32026.061.841.01
AdamW2026.062.012.01
Adafactor2026.063.011.01