Ling 3.1 flash released
- Source
- huggingface.co
- Author
- InclusionAI model releases
- Date
Context
Ling-3.1-flash's active parameter count describes the computation used for each token, not the size of the model that must be served. InclusionAI's model card describes a design that activates 8 of 512 routed experts per token, so 25 billion of its 560 billion parameters are active at a time. Its recommended server setups use eight H200 or four GB300 GPUs.
That distinction matters when comparing it with a smaller dense model, which uses all of its parameters for every token. Selecting only some experts can reduce per-token computation, while the full set of weights still has to reside somewhere in the serving system. The documented hardware setups illustrate substantial deployment requirements; the active count alone is not a reliable estimate of the memory needed to run it.
- 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.
- reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
Checking sign-in…
Loading comments…