AI2 releases text-generating AI models and their training data to the public

The Allen Institute for AI (AI2) is introducing its GenAI language models named OLMo, emphasizing their openness and flexibility for use in various contexts, including training, experimentation, and commercialization. Unlike many existing text-generating models developed by other organizations, which may be considered less open due to their proprietary training data, OLMo models are designed to be more accessible, accompanied by the code, data, and metrics used in their creation. Despite certain limitations such as lesser performance in non-English languages and code generation, the initiative represents a significant step towards enhancing transparency and collaboration in AI research. Moreover, the models, particularly the OLMo 7B, present a competitive alternative to existing models like Meta’s Llama 2 in specific benchmarks. AI2 is committed to further development, including the release of more advanced models and data sets, all shared freely to foster ongoing innovation and research in the field.