List your company

Language Modeling on FineWeb-edu 12 text passages (held-out)

1.683Average LossFull Qwen 3B
1.511882.666943.8224.97706Apr 2, 2026

Evaluation results

MethodLinks
Full Qwen 3BNumber of layers=362026.041.683——
17-layer baselineNumber of layers=172026.045.69—0
OUROBOROS V22026.045.96130