April 2024 Summaries

Snowflake Arctic is a groundbreaking open-source language model developed by Snowflake, boasting a size of 480 billion parameters, making it the largest open-source model currently available. It surpasses the performance of both Llama 3 8B and Llama 2 70B on various metrics while requiring less than half the training compute budget. Known for its proficiency in SQL and coding tasks, Arctic is offered under a liberal Apache 2.0 license. Users can easily run Arctic in the cloud using Replicate, with just one line of code, and explore its capabilities through an interactive API playground that provides adaptable code snippets in multiple languages. Additionally, Arctic can be operated using official client libraries for JavaScript and Python, as well as directly through cURL and other Replicate client libraries. Snowflake also encourages engagement with Arctic through a demo app, a hackathon, and community interactions via platforms like Twitter and Discord.

Llama 3, the latest language model from Meta, is now available on Replicate with a significant enhancement in performance and an 8000-token context window, doubling that of its predecessor, Llama 2. Users can run Llama 3 in the cloud with minimal setup, utilizing Replicate's API playground to experiment with prompts and explore the model's capabilities through interactive code snippets in various programming languages such as JavaScript, Python, and cURL. Llama 3 offers four model variants with different parameter sizes—70 billion and 8 billion—and comes in both base and chat-tuned forms, allowing users to choose the model best suited for their needs, whether focusing on accuracy or performance. For those interested in building chat applications, a demo chat app in Next.js is available for deployment on Vercel, with further customization options provided via a GitHub repository. Engaging with the community through platforms like Twitter and Discord is encouraged for updates and discussions about Llama 3 developments.