FP8
FP8 is a family of 8-bit floating-point formats for representing values and performing reduced-precision computation.
Build a DeepSeek Model (From Scratch)
An 8-bit floating-point numerical format used to represent and compute neural-network values with reduced precision.
- How it is used
- Uses FP8 quantization to reduce computation and memory costs in large-scale model training.
- How deep it goes
- Explains format trade-offs and the mixed-precision training framework, including scaling, accumulation, and forward and backward format choices.
- Examples
- Traces a linear layer’s mixed-precision data flow and compares E4M3 with E5M2.
- What you can do after
- Evaluate and understand FP8 precision choices and the scaling strategies used to support stable training.
Read how Build a DeepSeek Model (From Scratch) explains FP8 →
Quantization and Fast Inference
A family of 8-bit floating-point formats for reduced-precision numerical computation.
- How it is used
- Targets inference throughput and reduced model storage while also supporting mixed-precision training.
- How deep it goes
- Explains bit layouts, representable-value distributions, scaling choices, reconstruction error, and FP8 kernel constraints.
- Examples
- Compares OPT-6.7B weight reconstruction and measures weight-only storage and WikiText-2 perplexity.
- What you can do after
- Operate/configure FP8 by selecting formats and scaling strategies, then checking hardware and kernel requirements.
Also covered in
- Local Image GenerationIn early access (MEAP)
What to know first
- activationAn activation is a numerical value or tensor produced by a neural-network unit, layer, or operation during a forward computation.
- bfloat16bfloat16 is a 16-bit floating-point format with one sign bit, eight exponent bits, and seven fraction bits, providing a wide numerical range with reduced precision.
- forward passA forward pass is a computation that propagates inputs through a model or dependency structure in the forward direction to calculate outputs or intermediate values.
- optimizer stateOptimizer state is auxiliary information an optimization algorithm maintains across updates to determine or adapt its parameter updates.
Related concepts
- 8-bit signed integercontrasts with · An 8-bit signed integer is a fixed-width integer representation of whole numbers from −128 through 127.
- E4M3includes · E4M3 is an 8-bit floating-point format with one sign bit, four exponent bits, and three fraction bits.
- FP4contrasts with · FP4 is a family of four-bit floating-point formats for representing numerical values at reduced precision.
Questions about FP8
What is FP8?
FP8 is a family of 8-bit floating-point formats for representing values and performing reduced-precision computation.
Which Manning books explain FP8?
Build a DeepSeek Model (From Scratch) and Quantization and Fast Inference.
What should I know before learning FP8?
Manning books that teach FP8 usually introduce activation, bfloat16, forward pass and optimizer state first.
What is the difference between FP8 and 8-bit signed integer?
FP8 is a family of 8-bit floating-point formats for representing values and performing reduced-precision computation. An 8-bit signed integer is a fixed-width integer representation of whole numbers from −128 through 127.
What kinds of FP8 are there?
Kinds of FP8 include E4M3 and E5M2.
What is FP8 a kind of?
FP8 is a kind of low-bit floating-point format. A low-bit floating-point format is a floating-point representation that uses substantially fewer bits than conventional formats such as FP32.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.

