Granite 4.0 3B Vision: Compact Multimodal Intelligence for Enterprise Documents

Granite 4.0 3B Vision is a compact vision-language model for enterprise document understanding. It uses ChartNet, a million-scale dataset with 1.7 million chart samples across 24 types, enabling cross-modal chart understanding. The DeepStack Injection architecture routes abstract visual features to earlier layers for semantics and spatial features to later layers for detail. Deployed as a LoRA adapter on Granite 4.0 Micro, it supports both multimodal and text-only workloads. Achieves highest Chart2Summary score (86.4%), strong table extraction performance (92.1% TEDS on PubTablesV2), and 85.5% EM accuracy on semantic KVP extraction. Can operate standalone or integrate with Docling for end-to-end document processing.

This summary was automatically generated by AI based on the original article and may not be fully accurate.