Solar Mini 4
- Source
- upstage.ai
- Author
- Upstage editorial sitemap
- Date
Context
Solar Mini 4 uses only part of its model on each processing step, but a deployment still needs memory for the whole model. Upstage's release article describes 35 billion total parameters, with 3 billion active per token. Parameters are the model's learned numerical values; activating a subset reduces computation without making the stored model equally small.
Upstage explicitly says all 35 billion parameters must remain loaded. It reports that, for on-premises deployment, a version, which stores those values more compactly, can run on one H100 GPU with 80 GB of memory. That distinction matters when estimating hardware needs: the active parameter count describes computation per step, while the full model determines the weights that must fit.
- quantization — Shrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.
- mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…