About
The OLMo family of large language models (LLMs) is a series of open-source models developed by the Allen Institute for Artificial Intelligence (AI2) to advance the science of language modeling. These models stand out for their openness, offering researchers access to the training data, code, models, and evaluation resources. This transparency helps examine various aspects of LLM development, including biases and risks. Trained on the Dolma dataset, OLMo models utilize the Tulu SFT mixture and a refined UltraFeedback dataset for enhanced question answering. The family includes different models with varying parameters, such as 1B and 7B, reflecting distinct training phases and optimizations, and encourages collaborative research within the open-source AI community 123.