Evidence-backed signal

Model size alone was becoming an insufficient description of frontier progress; training-token allocation and inference cost emerged as important variables in the model-compute relationship.

supported inference 95% confidence increasing

Downward provenance

event

Chinchilla preprint reframes compute-optimal language-model scaling

Tue Mar 29 2022 13:38:03 GMT+0000 (Coordinated Universal Time)
98% confidence
artifact

Training Compute-Optimal Large Language Models

paper_preprint
art-deepmind-chinchilla-20220329
original source

Training Compute-Optimal Large Language Models

arXiv
src-arxiv-chinchilla-20220329