StableLM

Why we included this project

StableLM is Stability AI's own line of open language models, and this repository is where the actual checkpoints live. You'll find the 3B and 7B Alpha base models alongside StableVicuna-13B, an RLHF fine-tune of Vicuna, so you can grab a size and fine-tune it yourself instead of going through a hosted API. The 3B-4E1T model stands out because it was trained on repeated tokens to study data-constrained scaling, and the README spells out the training setup and benchmark results so you can judge whether a given size meets your latency and quality targets. If you're evaluating small, self-hostable models for a specific task, this repo gives you the weights and configs to make that call.