sensenova/SenseNova-SI-1.1-Qwen2.5-VL-3B
SenseNova-SI-1.1-Qwen2.5-VL-3B is a 3 billion parameter multimodal foundation model from the SenseNova-SI family, built upon the Qwen2.5-VL architecture. Developed by SenseNova, it is specifically designed to enhance spatial intelligence in multimodal tasks. This model excels across various spatial intelligence benchmarks, demonstrating robust performance in visual understanding and spatial reasoning.
Loading preview...
Overview of SenseNova-SI-1.1-Qwen2.5-VL-3B
SenseNova-SI-1.1-Qwen2.5-VL-3B is a 3 billion parameter multimodal foundation model developed by SenseNova, part of the broader SenseNova-SI family. This model is specifically engineered to address deficiencies in spatial intelligence within multimodal AI, building upon the established Qwen2.5-VL architecture. It has been systematically trained on SenseNova-SI-8M, a curated dataset of eight million diverse data samples focusing on spatial capabilities.
Key Capabilities
- Enhanced Spatial Intelligence: Demonstrates strong performance across a range of spatial intelligence benchmarks, including VSI-Bench (54.9%), MMSI (30.8%), MindCube (52.6%), ViewSpatial (43.5%), and SITE (37.8%).
- Multimodal Understanding: Maintains robust general multimodal understanding, achieving 84.9% on MMBench-En.
- Data-Driven Improvement: Benefits from training on a rigorously taxonomized dataset designed to cultivate spatial reasoning.
Good For
- Applications requiring advanced spatial reasoning and visual understanding.
- Research into scaling multimodal foundation models for improved spatial intelligence.
- Tasks where understanding object relationships, positions, and orientations in images is critical.