Expand description
2026-09-25: Nemotron-H weight structs and loaders for the Mamba-2 mixer, attention and MoE layers.
Owner: model-layers (weight loading). Invariants: none beyond the types.
Structs§
- Nemotron
Expert Weight - 2026-09-25: Nemotron-H expert:
up_projanddown_proj, nogate_proj. - Nemotron
MoeWeights - 2026-09-25: Nemotron-H MoE layer weights.
- Nemotron
SsmWeights - 2026-09-25: Nemotron-H Mamba-2 mixer weights.
Enums§
- Nemotron
SsmQuant - 2026-09-25: Format of a Mamba-2 layer’s
in_proj, detected at load.
Functions§
- load_
nemotron_ attention - 2026-09-25: Load a Nemotron-H attention mixer.
- load_
nemotron_ moe - 2026-09-25: Load a Nemotron-H MoE mixer.
- load_
nemotron_ ssm - 2026-09-25: Load a Nemotron-H Mamba-2 mixer (
{layer_prefix}.mixer).