Stochastic AI Xturing

What is Stochastic AI Xturing?

Xturing is an open-source library from Stochastic that simplifies building and fine-tuning large language models. It provides a straightforward way to customize LLMs and create datasets from your own data, supporting models such as LLaMA, GPT-J, GPT-2, OPT, Cerebras-GPT, Galactica, and Bloom. The library is approachable for newcomers while offering the flexibility experienced developers need.

The Stochastic team’s design goals for Xturing are simplicity, productivity, efficient use of compute and memory, and customizability. The library was built by a global team with expertise in AI research and engineering, machine learning, and computer science, with the mission of making AI more accessible.

What is Stochastic AI Xturing used for?

Xturing covers the core workflows of a fine-tuning pipeline:

• Fine-tune models: Tailor LLMs to your application’s requirements.
• Generate datasets: Create, prepare, and save datasets from your own data sources.
• Evaluate models: Assess how your modified models perform.
• Save and load models: Manage your fine-tuned models.
• Run inferences: Use models for predictions or generating outputs.
• Contribute: The project is open source, so you can take part in its development.

Documented fine-tuning examples include:

• LLaMA on the Alpaca dataset
• GPT-J, with or without INT8
• Cerebras-GPT, GPT-2, Galactica, OPT, and a Generic Wrapper LLM on the Alpaca dataset, with or without LoRA and INT8
• LLaMA 7B, LLaMA 2 7B, and Falcon 7B on the Alpaca dataset, with or without LoRA and INT8

Who is Stochastic AI Xturing for?

Xturing is aimed at developers of all levels, including anyone who wants to work with AI regardless of prior experience.