Mamba Architecture
Skill Verified ActiveState-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.
To provide users with a comprehensive understanding of the Mamba state-space model architecture, its advantages over Transformers, and practical guidance on its implementation and usage.
Features
- Detailed explanation of Mamba's Selective SSM
- Comparison of Mamba-1 vs Mamba-2
- Code examples for basic usage and language modeling
- Performance benchmarks (speed, memory, perplexity)
- Installation instructions and common issue resolution
Use Cases
- Understanding alternative sequence modeling architectures to Transformers
- Implementing or fine-tuning Mamba models for long-context tasks
- Evaluating Mamba's performance benefits for inference and training
- Debugging common issues encountered when working with Mamba
Non-Goals
- Providing a pre-trained Mamba model for direct use
- Implementing advanced Mamba features beyond basic configuration and usage
- Covering non-Python implementations of Mamba
Trust
- info:Issues AttentionThe repository has 17 open issues and 4 closed issues in the last 90 days, suggesting a moderate level of maintainer engagement but with a potentially slow closure rate.
Installation
npx skills add davila7/claude-code-templatesRuns the Vercel skills CLI (skills.sh) via npx — needs Node.js locally and at least one installed skills-compatible agent (Claude Code, Cursor, Codex, …). Assumes the repo follows the agentskills.io format.
Quality Score
VerifiedTrust Signals
Similar Extensions
Mamba Architecture
99State-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.
Rwkv Architecture
99RNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV cache. Train like GPT (parallel), infer like RNN (sequential). Linux Foundation AI project. Production at Windows, Office, NeMo. RWKV-7 (March 2025). Models up to 14B parameters.
Rwkv Architecture
96RNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV cache. Train like GPT (parallel), infer like RNN (sequential). Linux Foundation AI project. Production at Windows, Office, NeMo. RWKV-7 (March 2025). Models up to 14B parameters.
TorchTitan Distributed LLM Pretraining
99Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.
Implementing Llms Litgpt
98Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model implementations, educational understanding of architectures, or production fine-tuning with LoRA/QLoRA. Single-file implementations, no abstraction layers.
Distributed Llm Pretraining Torchtitan
98Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.