Understanding the Field with Our Own Eyes and Ears: The Development Story of Our Own Encoders and HyperCLOVA X SEED 4B

Team NAVER has developed HyperCLOVA X SEED 4B, a lightweight omni-modal language model optimized for environments that require real-time processing of diverse forms of information such as drone footage, satellite imagery, and radio audio. Unlike existing large-scale models, this model targets use in defense and industrial fields where security and real-time performance are critical.

The core of this development is that the encoders serving as the model's 'eyes' and 'ears' were completed with proprietary technology, without relying on external models.

• HyperCLOVA X CLIP: A 637M-scale vision encoder trained from-scratch without using external weights, with strengths in Korean OCR and understanding of Korean-specific context.
• Audio Encoder: Converts sound information such as dialogue, background sounds, and sound effects within video into tokens that the language model can understand.

As a lightweight 4B-scale model, HyperCLOVA X SEED 4B applies high-resolution image processing and audio token compression technology. Through this, it has achieved a design capability that allows it to operate efficiently without latency across various fields with limited computing resources, such as smartphones, robots, drones, and tactical vehicles.