The Definitive Guide toAI Data Centers
Ask the GuideAboutAccount
Guide › Glossary › BF16

BF16 · Brain Float 16

A 16-bit floating-point format with a wide exponent range, a common default for stable AI training.

Current numbers

~14 bytes/paramstated BF16 weights + FP32 master and two FP32 Adam moments; unique serialized state, before implementation metadata or replicationas of 2026 · register ↗
40 GiB aggregate; 5 GiB/rank at ideal TP=8Llama 3.1 70B logical BF16 KV at exactly 131,072 tokens; ideal eight-way head sharding, before overheadas of 2024-07 · register ↗
504 KiB/token (516.096 kB/token)Llama 3.1 405B aggregate logical BF16 KV per retained token, before layout or replicationas of 2024-07 · register ↗

← All terms