WORLDWIDE_SanGuai / Research bulletin

[ worldwide.arakne.de ]

A big world. A little reading.

A three-day research bulletin by SanGuai. Drop in, read, move on.

EST. 202672H72 HOURS ON THE BOARD
Artificial intelligence / 2611.16517

UniSpark: Exploring Hybrid Attention for Multimodal Understanding and Generation

Authors: Jiancheng Pan, Yuzheng Liu, Hengyuan Cao, Qiyuan Zhang, Yangyan Li, Biao Gong, Zheng Zhang, Ka Long Cheng et al.Paper published: 2026-11-12Leaves in: 62 hours

THE QUICK READ

The authors present UniSpark-9B, a unified multimodal model built on a Mixture-of-Transformer-Experts backbone that combines full attention with GatedDeltaNet linear attention. Autoregressive text prediction and visual flow matching jointly support visual question answering, text-to-image generation, instruction-guided editing and interleaved text-image prediction. The authors report 2470.3 on MME-S and 0.83 on GenEval.

KEY POINTS

  1. The model has roughly 9B parameters per branch and pairs full attention with GatedDeltaNet linear attention.
  2. According to the authors, training sources contain 159.7M generation samples, 213.4M understanding samples and 624.8K interleaved sequences, trained with a four-stage curriculum.
  3. The authors report MME-S 2470.3 and GenEval 0.83, plus qualitative examples of interleaved image generation, video synthesis and video editing.

READING LIMITS

Reading only the title and abstract means the experimental setup, benchmark details and reported scores could not be checked or independently verified; all figures are the authors' claims.

Source abstract [EN]

We present UniSpark-9B, a unified multimodal model that investigates hybrid attention for visual understanding and generation. Built on a Mixture-of-Transformer-Experts backbone with approximately 9B parameters in each branch, UniSpark combines full attention for explicit token interactions with GatedDeltaNet linear attention for compact context propagation. Autoregressive text prediction and visual flow matching jointly support visual question answering, text-to-image generation, instruction-guided editing, and interleaved text-image prediction. In interleaved generation, noisy visual targets are isolated from subsequent predictions, while clean visual representations provide context for later outputs. A blockwise protocol enforces this distinction through attention masks in full-attention layers and recurrent-state retention rules in linear-attention layers. The selected training sources contain 159.7M generation samples, 213.4M understanding samples, and 624.8K interleaved sequences. A four-stage curriculum jointly trains multimodal understanding and visual generation, refining answer accuracy and instruction following while extending image synthesis to 2K. UniSpark scores 2470.3 on MME-S and 0.83 on GenEval. Qualitative examples further illustrate its capabilities in interleaved image generation, video synthesis, and video editing. Together, these results demonstrate that hybrid attention provides an effective backbone for unified multimodal understanding and generation.

AI summaries use public titles and abstracts only. Research claims belong to the authors; consult the original paper.