State-space model with O(n) complexity vs Transformers' O(n²). 5× faster inference, million-token sequences, no KV cache. Selective SSM with hardware-aware design. Mamba-1 (d_state=16) and Mamba-2 (d_state=128, multi-head). Models 130M-2.8B on HuggingFace.
7.0
Rating
0
Installs
AI & LLM
Category
Excellent skill documentation for Mamba state-space models. The description clearly communicates the key value proposition (O(n) complexity, 5× faster inference, no KV cache). SKILL.md provides comprehensive, actionable workflows covering basic usage, pretrained models, Mamba-1 vs Mamba-2 comparison, and benchmarking. Code examples are complete and practical. Structure is clean with logical progression from quick start to advanced topics. The 'When to use vs alternatives' section effectively positions Mamba against Transformers and other architectures. Minor deductions: (1) novelty is moderate since it's primarily a wrapper around an existing library with well-documented APIs, though the architectural comparison adds value; (2) some referenced files in advanced topics aren't present in the directory tree (selective-ssm.md, mamba2-details.md, performance.md vs architecture-details.md, benchmarks.md, training-guide.md) suggesting potential misalignment, though per instructions this doesn't affect scoring. Overall, a CLI agent could confidently work with Mamba models using only this skill.
zechenzhangAGI
Skill Author
Loading SKILL.md…
