Expand description
2026-09-25: The generic transformer model, TransformerModel, and the served NLLB model.
Layer-specific logic lives behind the TransformerLayer trait objects in
TransformerModel::layers; trait_impl implements Model for it.
Owner: model-engine. Invariants: none beyond the types.
Modules§
- nllb
- 2026-09-25: Served NLLB-200 / M2M-100 encoder-decoder translation model.
Structs§
- Transformer
Model - 2026-09-25: One loaded model: its layers, weights, kernel handles, KV and
SSM pools and CUDA-graph caches.
trait_implimplementsModelfor it.