Expand description
Per-block affine quantization descriptor (scheme tag 0x03).
The tensor is divided into fixed-size contiguous blocks along a specified axis.
Each block carries its own scale (and optionally zero_point). Partial
trailing blocks are permitted; see docs/spec/quantization/per-block-affine.md
§ Padding for the normative padding rules.
See docs/spec/quantization/per-block-affine.md for the normative definition.
Structs§
- PerBlock
Affine - Quantization parameters for per-block affine quantization.
Constants§
- FLAG_
SYMMETRIC - Bit 0: symmetric flag — zero-point array is implicit (all zeros).
- PER_
BLOCK_ AFFINE_ MIN_ BLOCK_ SIZE - Minimum block size for per-block affine (must be a power of two ≥ 2).
- RESERVED_
MASK - All bits except bit 0 are reserved and must be zero.