AI engineering glossary · LLM training alignment
What is MFU (Model FLOPs utilisation)?
Model FLOPs utilisation (also called MFU): The share of a hardware system's peak theoretical FLOPs that a training job turns into useful model computation, excluding recomputation overhead. Values of 30-50 percent are considered good for large runs; low values signal communication or data bottlenecks.
Model FLOPs utilisation: an example
A cluster with 1,000 units of peak capacity delivering 400 units of model math has an MFU of 40 percent.
Related terms
More LLM training alignment terms
- Model merging
- Next-token prediction
- NormalFloat 4-bit
- Odds ratio preference optimisation
- Overtrained model
- Paged optimisers
- Parameter-efficient fine-tuning
- PEFT library
Where you learn it
Model FLOPs utilisation is part of the AI Engineering Program, which teaches it with notes, a hands-on lab and an exam. The first lesson is free.