Expand description
Mamba-1 model internals. Mamba-1 CPU runtime internals (model, kernels, tensors, weights).
Structsยง
- Config
- Mamba-1 model configuration.
- Full
Adam State - Adam moments for full-parameter online Mamba training.
- Model
- Mamba model weights and inference kernels.
- Scratch
Buffers - Preallocated temporary buffers for token forward passes.
- State
- Recurrent runtime state for a Mamba model.
- Train
Scope Mask - Train-scope mask for Mamba full-parameter online updates.