Explore skills

A comprehensive open-source library of AI research and engineering skills for any AI model. Package these skills to empower your Claude Code/Codex/Gemini agent as a full-horsepower AI research agent, maintained by Orchestra Research.

Provides guidance for experiment tracking with SwanLab. Use when you need open-source run tracking, local or self-hosted dashboards, and lightweight media logging for ML workflows.

Expert guidance for fast fine-tuning with Unsloth, offering 2-5x faster training, 50-80% less memory, and LoRA/QLoRA optimization.

A language-independent tokenizer that processes text as raw Unicode, supporting BPE and Unigram algorithms. It's fast, lightweight, and offers deterministic vocabulary, ideal for multilingual and CJK support, or reproducible tokenization.

Expert guidance for fine-tuning LLMs with LLaMA-Factory, featuring a no-code WebUI, over 100 models, 2/3/4/5/6/8-bit QLoRA, and multimodal support.

Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activation caching. Use when reverse-engineering model algorithms, studying attention patterns, or performing activation patching experiments.

Simple Preference Optimization for LLM alignment. This reference-free alternative to DPO offers better performance (+6.4 points on AlpacaEval 2.0) and enables simpler, faster training than DPO/PPO.

Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infrastructure from algorithms. Use for clean RL abstractions, easy algorithm experimentation, or scalable training with Monarch and TorchTitan.

An open-source embedding database for AI applications, offering storage for embeddings and metadata, vector/full-text search, and metadata filtering. It scales from notebooks to production and is ideal for semantic search, RAG, and local development.

Build complex AI systems with declarative programming, optimize prompts automatically, and create modular RAG systems and agents using DSPy, Stanford NLP's framework for systematic LM programming.

A multi-agent orchestration framework for autonomous AI collaboration, ideal for building teams of specialized agents on complex tasks, role-based collaboration with memory, or production workflows requiring sequential/hierarchical execution. It's built without LangChain dependencies for lean, fast execution.

A fully managed, auto-scaling vector database for production AI applications, offering hybrid search, metadata filtering, and namespaces with low latency. Ideal for RAG, recommendation systems, or semantic search at scale, especially with serverless infrastructure.