ModelRefs / BF16 (Brain Floating Point 16) — AI Glossary
BF16 (Brain Floating Point 16) — AI Glossary
A 16-bit floating point format with the same 8-bit exponent as float32 but 7 mantissa bits; the standard precision for LLM training.
Overview
BF16 was introduced by Google Brain for TPU training. Its float32-compatible exponent range prevents overflow/underflow during training without loss scaling. Most frontier model training uses BF16 mixed-precision: weights in BF16, optimizer states in FP32. Supported natively by A100, H100, and all modern AI accelerators.
Reference details
| Topic | training |
|---|---|
| Also known as | bfloat16, brain float 16 |
| Last reviewed | 2026-06-24 |
Related terms
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to BF16 (Brain Floating Point 16) — AI Glossary.
Frequently asked questions
What is BF16 (Brain Floating Point 16)?
A 16-bit floating point format with the same 8-bit exponent as float32 but 7 mantissa bits; the standard precision for LLM training.
Is BF16 (Brain Floating Point 16) the same as bfloat16?
Yes — bfloat16, brain float 16 are common aliases for BF16 (Brain Floating Point 16).
What concepts are related to BF16 (Brain Floating Point 16)?
Closely related concepts include mixed precision, quantization, h100.