on February 3, 2024
Hot to Follow
OLMo Unveiled: Exploring the Unmatched Capabilities of a Truly Open and State-of-the-Art Language Model and Framework
The Allen Institute for AI (AI2) has just released its first Open Language Model (OLMo), a 7B parameter model. It includes open training code, open data, full model weights, evaluation code, and fine-tuning code. It shows strong performance on many generative tasks. There is also a smaller version, OLMo 1B. The collaborative effort from AMD, CSC – IT Center for Science (Finland), Mosaic/Databricks, Kempner Institute at Harvard University, and the University of Washington. Additional with EleutherAI, Meta, Stanford CRFM, TogetherAI and HuggingFace.The Open Language Model (OLMo) is an AI2 LLM framework intentionally designed to provide access to data, training code, models, and evaluation code necessary to advance AI through open research. This empowers academics and researchers to collectively study the science of language models.OLMo and the framework include full pretraining data. The model is based on AI2’s Dolma dataset, which includes a three trillion token open corpus for language model pretraining, along with the code that generates the training data.The OLMo framework provides complete model weights for four model variants at the 7B scale, each trained on at least 2T tokens. Inference code, training metrics, and training logs are all included. Evaluation:We have released the evaluation suite used in development, which includes 500+ checkpoints per model taken every 1000 steps during the training process, and evaluation code under the Catwalk project. Each model is accompanied by the full training data used, including the code that produces the training data, from AI2’s Dolma and WIMBD for analyzing pretraining data. Full model weights, training code, training logs, training metrics in the form of Weights & Biases logs, and inference code are available. There are over 500 checkpoints per model, taken from every 1000 steps during the training process, which are available as revisions on HuggingFace.
Evaluation code is available under the umbrella of AI2’s Catwalk and Paloma. Fine-tuning code and adapted models will be available soon with Open Instruct. All code, weights, and intermediate checkpoints are released under the Apache 2.0 License.
We are going to show you how to prompt, so watch this space!
Fun & Curious
Stewart’s Ice Cream will release a new flavor this month to honor RPI’s 200th anniversary
Stewart’s Shops has renamed a flavor in honor of Rensselaer Polytechnic Institute’s bicentennial. Quantum Freeze, a vanilla ice cream with a tart cherry swirl and blue and red Pop Rocks, will be the new flavor in the store’s ice cream lineup. “We are excited to announce that this flavor was inspired by the cutting-edge cooling system of the university’s IBM Quantum System One supercomputer. Last November, we received over 500 potential ice cream flavor names from students, faculty, staff, and alumni, and Quantum Freeze was the clear winner of the final vote!” The school is celebrating its 200th anniversary with a variety of events, such as a new ice cream flavor, a bicentennial Big Red Freakout hockey game, and an evening with U.S. Poet Laureate Tracey K. Smith, as stated in the release.
Important Events
Did you know that AI on Pi Day is happening again on March 14th, 2024? Don’t miss out on this amazing opportunity!
This year’s flagship AIDA User Group conference is going to be bigger and better than ever, featuring a full 24 hours of AI! AI on Pi Day is the perfect opportunity to learn about AI, data, and analytics through a series of brief and engaging talks (30 minutes with Q&A). It’s completely free for all AIDA User Group members (and membership is also free!).
The best part? The call for papers (CfP) is currently open at https://aidaug.org/pidaycfp. Please submit your application and invite your friends to do the same if you want to be a part of our story!
Previous editions have featured incredible speakers, including Seth Dobrin, Bill Higgins, and Rob Thomas.
We are thrilled to announce that the program will be launching soon, so take a part!
AIDA’s Best Ever AI Game Show at IBM TechXchange EMEA
At IBM TechXchange EMEA in Barcelona we created first AI Game Show Ever! Ready to take a look?!
We work on:
Bitol is a Linux Foundation AI & Data Sandbox project licensed under the Apache 2.0 license. As of now, it defines an open standard for data contracts called Open Data Contract Standard.
Share and Stay Tuned
Become a Member if you want to be a part of the story.
Thank you for sharing your stories! Feel free to share Events and News you find interesting with us here! We will give it a shout!
For exciting updates and valuable insights, visit us at aidausergroup.org and on LinkedIn. Stay tuned for more!