[KongchangAI]
ModelGlider 2.5

Glider 2.5

A series of lightweight open-source language models released by Fastino AI, with parameter counts ranging from 74 million to 300 million. The models are designed for local deployment on consumer-grade hardware and low-latency inference, and are suited to tasks such as guardrails and real-time decision-making.

Timeline (last 90 days)

Oct 29

Fastino AI released the Glider 2.5 model.

Unverified50%
Oct 29

Glider 2.5 is a model with only 74 million parameters.

Unverified50%
Oct 29

The Glider 2.5 series covers a range from 74 million to 300 million parameters.

Unverified50%
Oct 29

On the guardrail benchmark, Glider 2.5 performed better than Jev AI in at least several scenarios.

Unverified50%
Oct 29

The 74-million-parameter Glider model was used to play the game Doom in tests.

Unverified50%
Oct 29

The Glider 2.5 model is very small and easily runs on consumer-grade hardware.

Unverified50%
Oct 29

A 74-million-parameter model can achieve high throughput on a consumer-grade CPU, making an end-to-end real-time control chain possible without offloading inference to the cloud.

Unverified50%
Oct 29

The video test is a preliminary observation from a single source, with limited sample size and test scope, and a complete evaluation on one's own data distribution is needed before adoption in a production environment.

Unverified50%

All Facts (8)

Source Articles