OLMoTrace is powered by my previous work infini-gram, with some innovative algorithmic twists. Really proud to turn an academic research project into a real LLM product, it’s been a truly amazing experience.

Check out infini-gram: infini-gram.io

Try OLMoTrace in Ai2 Playground with our OLMo 2 models: playground.allenai.org

If OLMoTrace gives you new insight into how LLMs behave, we’d love you to share your use case! 💡Take a screenshot, post the thread link if you like, and don’t forget to tag
@allen_ai

(4/4) Searching in OLMo 2's training data is now available in both our web interface and the API endpoint.

Plus, OLMo 2 32B Instruct is a very strong model. Let's do real science with it 🧪

(3/4) We know the pain point in LLM research in academia: We don't know what's in the training data of these LLMs (GPT, Llama, etc) and what's not; we can only speculate.

So we made the full training data of OLMo 2 and OLMoE searchable, including pre-training and post-training.

(2/4) Check out the source code of infini-gram here: github.com/liujch1998/infini-gram

If you are new to infini-gram, you might want to start with exploring our web interface infini-gram.io/demo and API endpoint infini-gram.io/api_doc

Yes we’ve read your paper and there’s so many interesting findings! Let’s grab coffee at Neurips