Expand description
2026-09-30: --activation-quantization: which activation precision each decode projection
runs at, as a function of the projection family and of the number of rows in the launch
(decode batch rows, MTP verify rows included).
A value is either one format for every row count, or a ladder that maps row ranges to formats, optionally overridden per projection family:
spec := ladder ("," family ":" ladder)*
ladder := format | rung (";" rung)*
rung := range "=" format
range := N | N "-" | N "-" M (1-based rows, inclusive)
format := "bf16" | "fp8" | "nvfp4" | "declared" | "adaptive"
family := "gdn" | "attn" | "ffn" | "moe" | "lm_head"bf16,fp8andnvfp4run that activation format through one fixed-order kernel family per weight format, so a row’s output bits do not depend on the row count or on the other rows.declaredis the checkpoint’s declared input-activation format per layer (quantization_config), run the same way.adaptiveis today’s routing: each dispatch site picks its kernel by row count as it did before this flag, byte for byte. The named valueadaptiveis the ladder1-=adaptive; a rung may also hand only its range to it (1=bf16;2-=adaptive).- A ladder’s rungs start at row 1, leave no gap and do not overlap, and the last one is open-ended. A family override replaces the base ladder for that family only.
The same shape, a rule per (family, row range), is what the auto-fuser’s circuit rules express per row range (kernels/circuits); a later step makes the circuit read this value.
Owner: config (quantization). Invariants:
- Pure: no environment, no I/O, no process state. The serve parses the flag here once and
publishes the result; every dispatch site asks
ActivationQuantization::route. - Its
Displayform is canonical and parses back to an equal value.
Structs§
- Activation
Quantization - 2026-09-30: A parsed
--activation-quantizationvalue. - Ladder
- 2026-09-30: Rungs covering every row count from 1, in order, without gaps or overlaps.
- Rung
- 2026-09-30: One rung: rows
lo..=hi(hiNone: no upper bound) runformat.
Enums§
- ActQuant
Format - 2026-09-30: The activation format of one rung.
- Proj
Family - 2026-09-30: The projection families a family override can name.