List your company

Language Modeling on Wikipedia English (BPE 16k, context 256) (val)

16.69Validation Perplexity (PPL)D=5 concat
16.2719.10521.9424.775Jun 25, 2026

Evaluation results

MethodLinks
D=5 concatC=768, FLOPs/tok=42.5M2026.0616.69—
Roformer N=6C=768, FLOPs/tok=42.5M2026.0617.951.26
D=6 addC=446, FLOPs/tok=15.9M2026.0620.4—
Roformer-hFFN N=6C=446, FLOPs/tok=15.9M2026.0621.441.04
D=2 concatC=446, FLOPs/tok=7.2M2026.0625.48—
Roformer N=3C=446, FLOPs/tok=7.2M2026.0627.191.71