Initially, it seemed as though "big AI😆" and its gated LLMs would dominate, but it's been inspiring to witness the open-source community's retaliation. While models like LLaMA, Alpaca, Vicuna, Koala, and GPT4All are impressive, their licenses prevent commercial usage, rendering them unusable for businesses. Fortunately, there are now options such as Dolly 2, StableLM, and Cerebras GPT that businesses can employ to expand their AI capabilities without compromising data privacy and security.
The work done by Databricks is an excellent example of how the field can progress, as they have made Dolly 2.0's model weights, training code, and complete dataset available for commercial use (unlike Dolly v1). Cerebras Systems has also released a suite of new models under an Apache 2 license, called CerebrasGPT. Additionally, Stability AI has published the StableLM Base Alpha models under a commercial license, although it's worth noting that the fine-tuned checkpoints can only be used for research purposes, as they are partially trained on data from Stanford's Alpaca, which prohibits commercial usage.
https://vicuna.lmsys.org/
https://github.com/nomic-ai/gpt4all
https://crfm.stanford.edu/2023/03/13/alpaca.html
https://en.wikipedia.org/wiki/LLaMA
Comments
Post a Comment