SenseNova-Loop-9B-Preview: 9B parameters, looped transformer architecture

SenseNova-Loop-9B-Preview is a multimodal language model that processes images and text simultaneously. It performs text generation and image understanding within a single pipeline and can be called directly via the image-text-to-text pipeline in the Transformers library.

1 / 8

The model's most notable feature is its adoption of the 'looped transformer' architecture. It has a total of 9.65 billion (approx. 9.6B) parameters and is released under the Apache 2.0 license, allowing free use in commercial projects. It supports English and Chinese.

It is compatible with major inference servers such as vLLM and SGLang, making it easy to serve via an OpenAI-compatible API. Because the model includes custom code, the trust_remote_code option must be enabled. It can also be run immediately in cloud environments such as Colab or Kaggle.

With a lighter 9B scale compared to existing large multimodal models, it is suitable for developers who want to efficiently test multimodal inference on local GPUs or cloud instances. It covers a wide range of scenarios, from complex vision tasks to interactive text generation.