[Lane 1 · numerics] GGUF IQ dequant — llama.cpp reference blocks and grid tables #1409

Sub-issue of #1408 ([Feature]: GGUF dequantization for the IQ1/IQ2/IQ3 families and MXFP4).

Lane: 1 · Numerics/Research
Skill needed: C reading; Python. No Kotlin needed.
Size: s (a few hours)
Blocked by: nothing

What to do

1. From llama.cpp ggml/src/ggml-quants.c and ggml-common.h, document the block layout (bytes per block, field order) of each type listed in the parent and the grid tables each one needs.
2. Produce one small reference tensor per type (e.g. with gguf-py or llama.cpp's quantize tool) plus its dequantized float values, and attach them here.

Acceptance

• Layout notes for every type
• Reference data per type attached
• Result reported back on the parent issue

Notes

If something here turns out to be wrong or much bigger than the size label, comment on this issue or open a specific bug instead of blocking the parent.

Coverage matrix: docs/modules/ROOT/pages/reference/ops-status-matrix.adoc · How lanes work: Issue taxonomy