Dolly
About
The Dolly family of large language models (LLMs), developed by Databricks, includes notable models like Dolly v1 and Dolly v2. Dolly v1, based on EleutherAI's GPT-J with 6 billion parameters, showcased that older, open-source models can exhibit strong instruction-following capabilities with limited fine-tuning on a high-quality dataset 248. Initially, its commercial use was limited by the licensing of its training data 48. To overcome this, Dolly v2 introduced a new dataset, "databricks-dolly-15k," which allows for both research and commercial utilization, and includes a model with 12 billion parameters based on EleutherAI's Pythia-12b 13. The Dolly models are engineered to comprehend and execute instructions articulated in natural language. Although they may not match the cutting-edge models in terms of performance, they provide an economical and versatile solution for entities aspiring to develop customized LLMs 212.