Latest research
November 21, 2024
Tülu 3 opens language model post-training up to more tasks and more people
Tülu 3 is a leading instruction following model family, offering fully open-source data, code, and recipes.November 21, 2024
Tülu 3: The next era in open post-training
A technical deep-dive into Tülu 3, with the model "recipe", data, and more.November 19, 2024
Scientific literature synthesis with retrieval-augmented language models
Ai2’s & UW’s new retrieval-augmented LM helps scientists navigate and synthesize scientific literature.November 12, 2024
How many Van Goghs does it take to Van Gogh? Finding the imitation threshold
Meet MIMETIC^2: Finding the number of images required by text-to-image models for imitation of a concept.October 28, 2024
Hybrid preferences: Learning to route instances for human vs. AI feedback
We introduce a routing framework that combines inputs from humans and LMs to achieve better annotation quality.October 2, 2024
Investigating pretraining dynamics and stability with OLMo checkpoints
We use data from our open pretraining runs to test hypotheses about training dynamics in OLMo checkpoints.September 4, 2024
OLMoE: An open, small, and state-of-the-art mixture-of-experts model
Introducing OLMoE, the first model to be on the Pareto frontier of performance and size, released with open data.August 12, 2024