On March 11th, Elon Musk said xAI would open source its AI chatbot Grok, and now an open release is available on GitHub. This will allow researchers and developers to build on the model and impact how xAI updates Grok in the future as it competes with rival tech from OpenAI, Meta, Google, and others.
A company blog post explains that this open release includes the “base model weights and network architecture” of the “314 billion parameter Mixture-of-Experts model, Grok-1.” It continues saying the model is from a checkpoint last October and hasn’t undergone fine-tuning “for any specific application, such as dialogue.”
As VentureBeat notes, it’s being released under the Apache 2.0 license that enables commercial use but doesn’t include the data used to train it or connections to X for real-time data. xAI said in a November 2023 post that the LLM Grok was “developed over the last four months” and is targeted for uses around coding generation, creative writing, and answering questions.
░W░E░I░G░H░T░S░I░N░B░I░O░
— Grok (@grok) March 17, 2024
After Musk bought Twitter (now X), the code behind its algorithms was eventually released, and Musk has openly criticized companies that don’t open-source their AI model. That includes OpenAI, which he helped found but is now suing, alleging the company breached an original founding agreement that it would be open source.
Companies have released open-source or limited open-source models to get feedback from other researchers on how to improve them. While there are many fully open-source AI foundation models like Mistral and Falcon, the most widely used models are either closed-sourced or offer a limited open license. Meta’s Llama 2, for example, gives its research away for free but makes customers with 700 million daily users pay a fee and won’t let developers iterate on top of Llama 2.
When the Grok chatbot launched, accessing it required an X subscription (aka a paid blue check). Grok’s pitch was to be a more irreverent, up-to-date chatbot alternative to OpenAI’s ChatGPT or Google’s Gemini. Instead, in our early testing, it was unfunny and lacked anything that could help it stand out against the more powerful and advanced chatbots available elsewhere.
Very nice to see Grok go open-source & open-weights: https://t.co/RTlvQarPzy
— François Chollet (@fchollet) March 17, 2024
here's your DEEP DIVE into @grok's architecture!
— Andrew Kean Gao (@itsandrewgao) March 17, 2024
I just went through the https://t.co/8Y5cjeImg6, for this 314B open source behemoth with *no strings attached*.
pic.twitter.com/CraHKGqILe
Grok weights are out under Apache 2.0: https://t.co/9K4IfarqXK
— Sebastian Raschka (@rasbt) March 17, 2024
It's more open source than other open weights models, which usual come with usage restrictions.
It's less open source than Pythia, Bloom, and OLMo, which come with training code and reproducible datasets. https://t.co/kxu2anrNiP pic.twitter.com/UeNew30Lzn
Aucun commentaire:
Enregistrer un commentaire