mamba-architecture - Mamba Architecture and Selective State Space Models
Provides workflows for configuring, loading, and generating text with Mamba selective state-space models.
Tags
Updated: 2026-09-18Capabilities
Typical Inputs
What this skill does
- Configure Mamba and Mamba2 models
- Load pretrained HuggingFace Mamba models
- Generate text with Mamba models
- Benchmark inference speed against Transformers
- Enable gradient checkpointing for optimization
Inputs
- Model configuration parameters
- Input tensor or token IDs
- Pretrained model checkpoint name
- Text prompt
Outputs
- Output tensor
- Generated text sequence
- Inference benchmark metrics
Requirements
- Linux OS
- NVIDIA GPU with CUDA 11.6+
- PyTorch 1.12+
- mamba-ssm Python package
- causal-conv1d Python package
