88Z AI, 14 September — Following the open-source release of its 14B main model, Ant Lingbo Technology has launched three additional models for the LingBot-World 2.0 series. These include a Small version for consumer GPUs, a bidirectional Teacher model for high-quality distillation, and a causal pre-training base supporting multi-modal inputs, aiming to lower deployment barriers and improve training efficiency.
At a glance
- New 1.3B Small model supports real-time local generation on consumer single GPUs
- Release of bidirectional Teacher model and causal pre-training base optimizes distillation efficiency and input compatibility
The update
Ant Lingbo Technology continues to enhance the LingBot-World 2.0 open-source ecosystem. After releasing the 14B main model in July, it has recently introduced three complementary models. The Small model, with 1.3B parameters, is designed for consumer-grade single GPUs and supports real-time world generation locally. The Bidirectional model serves as a two-way Teacher model and a high-quality distillation source. By combining consistency distillation with distribution matching distillation, it facilitates the training of lighter and faster Student models. Additionally, Causal Pretrain provides a causal pre-training foundation that supports both camera pose and text prompt inputs, offering flexible infrastructure support for developers.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 蚂蚁灵波科技Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.