Stability.AI & StableLM
Created on April 24|Last edited on April 24
Comment
Stability.AI, the company behind Stable Diffusion, just released StableLM. A part of their larger plan to release foundation models across all domains. Their models are much smaller than GPT-3, ranging from 3B to 7B (and 15B and 30B on the way) parameters. They aim for these models to be accessible and serve as the foundation for future research, very akin to Meta's LLaMA models. What makes these models great is that they're hosted on HuggingFace Spaces and a programmatic interface is available!
Curious, I tested StableLM-7B on an SAT question from The Princeton Review. I tested it alongside Bing and ChatGPT. Below is the question for reference.
StableLM's response:
ChatGPT's response:
Bing's response (in precise mode):
Though StableLM does have remarkable generative capabilities, it does fall a bit short when it comes to answering SAT math questions at least! Granted this model is much smaller than GPT-3 and Bing's GPT-4-powered AI. For its size, it does perform well. Future models (and possibly larger models) by Stability will undoubtedly generate better results.
References
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.