johnma2006/mamba-minimal

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

2965 stars · 218 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A single-file, readable PyTorch implementation of the Mamba selective state space model architecture. It produces numerically equivalent outputs to the official optimized implementation and can load pretrained Mamba checkpoints for text generation.

Use cases

• understand how the mamba ssm architecture works internally
• run a minimal mamba model in pure pytorch without custom kernels
• load pretrained mamba weights and generate text
• study annotated code for selective state space models
• prototype modifications to the mamba architecture
• compare mamba implementation against the official one numerically

When to choose

• you want to read and learn the mamba architecture from clean annotated code
• you need a dependency-light pytorch reference implementation for experimentation
• you want to verify correctness against the official implementation
• you are teaching or studying state space models

When to avoid

• you need fast, production-grade inference speed
• you require proper parameter initialization for training from scratch
• you want optimized CUDA kernels or long-context performance
• you need a maintained, feature-complete framework

Facets

library · maturity maintenance

deep-learning machine-learning llm-inference deep-learning large-language-models machine-learning python mamba ssm state-space-models educational minimal-implementation pytorch

Member repositories

For agents

markdown · JSON · MCP: product_card(name="johnma2006/mamba-minimal")