What is Petals?

Petal is an innovative decentralized platform that utilizes a powerful language model called Bloom-176b. Unlike traditional language models, Petal has the ability to load and utilize specific sections of the model for inference and fine-tuning purposes. With a remarkable speed of approximately 1 second per step (token) for single-batch inference, Petal can even perform parallel inference at a rate of hundreds of tokens per second.

What sets Petal apart from other language model APIs is its incorporation of fine-tune sampling techniques, enabling users to execute custom paths and gain insights into hidden states. Additionally, Petal provides a flexible PyTorch API, enhancing its usability and adaptability. It is worth mentioning that Petal is an integral part of the esteemed BigScience research workshop project.

Petals Use Cases

• Generating text.
• Processing natural language.
• Analyzing sentiment.