The company behind Stable Diffusion, a text-to-image AI, has now released StableLM, a so-called LLM (Large Language Model) that others can use to build, for example, ChatGPT competitors.
StableLM Compared to GPT-3 and GPT-4
StableLM is available in alpha form in two versions containing 3 and 7 billion parameters, respectively. They plan to release new versions of StableLM in the future with 15 and 65 billion parameters. This can be compared to GPT-3 and GPT-4, the language models used for ChatGPT, which have 175 billion and one trillion parameters, respectively.
StableLM can be downloaded via GitHub for those who want to start experimenting with the language model. It is released under the Creative Commons license BY-SA-4.0. This means that those who use StableLM must mention that Stability AI has developed the language model, and any products or services created using the language model must be provided with the same Creative Common license.
Other smaller language models
StableLM is just one of the “smaller” language models that have emerged in recent times. Among other options available are LLaMA (65 billion parameters) from Meta, Stanford Alpaca (7 billion parameters), Cerebras-GPT (111 million to 13 billion parameters), and Dolly 2.0 (2.8 billion parameters).
Above and below are two examples of what a ChatGPT-like interface can achieve by using the StableLM language model.
