Claude Skill MartBrowse skillsQuick linesLearn by videoTerminal guideWhat is a Skill?
 
← Back to list

Mamba Architecture (Selective State Space Models)

Hands-on skill for installing, running and benchmarking Mamba-1/Mamba-2 linear-complexity SSMs.

Dev & CodingAdvanced★ 11,926⑂ 867AI score 8/10Last updated: Jun 16, 2026

What it does

Helps you actually run Mamba, a state space model with O(n) complexity instead of the Transformer's O(n²) attention.

  • Install steps for mamba-ssm / causal-conv1d plus prerequisites (Linux, NVIDIA GPU, CUDA 11.6+)
  • Code for Mamba blocks and full LMs via MambaLMHeadModel, including text generation
  • Loading pretrained HuggingFace checkpoints (state-spaces/mamba-130m … mamba-2.8b)
  • Mamba-1 (d_state=16) vs Mamba-2 (d_state=128, multi-head, RMSNorm, tensor parallelism) comparison
  • Benchmark commands against a Transformer baseline (Pythia) and a per-model VRAM table
  • Fixes for common issues: OOM, slow builds, HuggingFace loading errors

Who it's for

  • ML engineers needing very long contexts (100K+ tokens) or streaming inference
  • LLM serving teams trying to cut KV-cache memory
  • Researchers comparing Transformer alternatives (RWKV, RetNet, Hyena)

Examples

  1. "Load mamba-2.8b on GPU and generate from this prompt" → gives pretrained loading + generate code
  2. "Swap my block for Mamba-2" → explains headdim, ngroups and the differences from Mamba-1
  3. "Measure whether Mamba beats Pythia on generation speed" → provides matched benchmark commands

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/Orchestra-Research/AI-Research-SKILLs/HEAD/01-model-architecture/mamba/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the 01-model-architecture/mamba folder from the GitHub repo Orchestra-Research/AI-Research-SKILLs into my ~/.claude/skills/mamba-architecture/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/Orchestra-Research/AI-Research-SKILLs.git /tmp/ai-research-skills && mkdir -p ~/.claude/skills/mamba-architecture && cp -r /tmp/ai-research-skills/01-model-architecture/mamba/* ~/.claude/skills/mamba-architecture/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open a terminal (Linux with an NVIDIA GPU recommended).
  2. Clone the skill repo: git clone https://github.com/Orchestra-Research/AI-Research-SKILLs.git
  3. Create the skill folder: mkdir -p ~/.claude/skills/mamba-architecture
  4. Copy the files: cp -r AI-Research-SKILLs/01-model-architecture/mamba/* ~/.claude/skills/mamba-architecture/
  5. Restart Claude Code and ask something like "use Mamba for a long-context model" to confirm the skill loads.
  6. Before running real code, install the packages: pip install causal-conv1d>=1.4.0 && pip install mamba-ssm --no-build-isolation
  7. Verify CUDA 11.6+ / PyTorch 1.12+ and available VRAM (the 2.8B model needs ~28GB in FP16).
View source on GitHub ↗License: MIT