Dolly

Dolly is Databricks’ open large language model that made history in early 2023 as the first openly released, instruction-following LLM to show ChatGPT-like behavior — proving a small team could create one in weeks rather than months. Dolly v2, a 12-billion-parameter model, was released with fully open weights and training data for commercial use.

Overview

Dolly v2 is based on EleutherAI’s Pythia-12B and fine-tuned on databricks-dolly-15k, a dataset of roughly 15,000 instruction-response pairs written entirely by Databricks employees — covering brainstorming, classification, question answering, summarization, and more. Because the data was human-generated rather than scraped from a proprietary chatbot’s outputs, Databricks could license the model and dataset openly, sidestepping the legal gray zone that surrounded many early open chatbots. The project became a landmark of the open-model movement and remains a popular baseline and teaching example. It has more than 10,800 GitHub stars; the repository was last updated in June 2023 and now stands as a stable, archived-in-spirit reference release rather than an actively evolving model line.

Key Features

• 12B-parameter instruction-tuned LLM with genuinely open licensing
• Trained on 15,000 human-written instruction examples, openly released
• Exhibits reasoning and instruction following far beyond its base model
• Runs on a single high-memory GPU for inference and fine-tuning
• Includes training code built on standard open tooling
• Historically important reference for open, commercially usable LLMs

💳 Pricing

Dolly is free and open source: Dolly v2’s weights are Apache-2.0 licensed for commercial use, and the dolly-15k dataset is shared openly as well.

Who Is It For

Developers and researchers who want a historically significant, openly licensed instruction model for experimentation, education, and baseline comparisons.