Making a small language model behave like a Mythos
https://tansei.io/blog/small-language-model-like-mythos
What the biggest AI models do better and simple ways to get similar results from a smaller model.
What the biggest AI models do better and simple ways to get similar results from a smaller model.
A live example of the idea: Meta distilled its open Muse Glimmer (~30B, Apache 2.0) from the much larger Muse Spark and aimed it specifically at local agents rather than general chat. The community's local results, quants and fine-tunes (including a GUI-grounding LoRA that outputs bounding boxes from a screenshot) are collected here if you want real-world data points for the article: https://shipwithmuse.live/categories/local-and-open-models (I help curate it)