Stability AI launches the Stable LM 3B language model, which claims to "bring high-performance experience to mobile platform devices"
Thanks to CTOnews.com netizen Alejandro86 for the clue delivery! CTOnews.com, October 3, Stability AI released an announcement yesterday, introducing a language model called Stable LM 3B, which is said to be suitable for mobile platform devices and "can bring a sustainable, high-performance experience to related devices."
▲ source Stability AICTOnews.com learned from the announcement that the Stable LM 3B model contains 3 billion parameters, focuses on text generation, has an autoregressive system based on Transformer decoder architecture, and uses a number of open source large-scale data sets for training.
▲ image source Stability AI officially said that the model uses 3 billion parameters and 256 NVIDIA A100 40GB GPU training. Although the parameters are less than similar large models, the performance is still impressive, and because of the smaller size and lower power consumption of the model, it is more suitable for mobile platforms.
In addition, the model has multi-platform compatibility and allows fine-tuning according to specific needs. At present, the model is open source on the Hugging Face platform, which is convenient for developers to use and improve. Interested friends can check it out.