liveBook · Manning

Range

As explained in Quantization and Fast Inference

Range is the interval of numerical values that a representation can encode.

Range is the interval of values a numerical representation can encode. With a fixed bit width, expanding that interval leaves fewer representable steps within it; narrowing the interval allows finer resolution. Quantization therefore requires choosing which values matter. section 1.3.2 section 2.1

For fixed-width integer representations:

  • An 8-bit signed integer has 256 possible values, from −128 to +127; values outside the chosen range cannot be represented. section 1.3.2
  • In Q3.4, the scale is 1/16, so the signed 8-bit range is −8 to 7.9375, with steps of 0.0625. section 2.1
  • For an 8-bit container, allocating more bits to the integer part doubles the range while halving precision. section 2.1

Q3.4 saturation

The value 10.0 exceeds the signed Q3.4 maximum of 7.9375, so it is stored as 7.9375. The fixed range prevents the original value from being represented. section 2.1

Figure 1.3 Floating point concentrates precision near zero (top), leaving large values sparsely represented. Integers use uniform spacing across your chosen range (bottom). Quantization is the act of deciding where to place that fixed grid.

Compare floating-point spacing, concentrated near zero, with the uniform grid used by integers across a chosen range. section 1.3.2

How Quantization and Fast Inference covers it

How deep it goes
Works through Q3.4 scaling and signed 8-bit limits, then illustrates saturation and bit allocation without developing calibration procedures.
Examples
Uses INT8 limits, Q3.4 encodings, and an input of 10.0 that saturates at 7.9375.
What you can do after
Evaluate how a fixed-point bit allocation trades representable range against step size.
How it connects
Connects range to quantization, resolution, fixed-point representation, floating-point representation, and bit width.

Also explained in

Compare how 2 books explain range →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Quantization and Fast Inference. The section links point to the parts of the book it draws on.