WikiofAuto
98.9% Line-Item Table & Bounding-Box Extraction Accuracy on Skewed ScansMethod: Multimodal LoRA on 60k Multi-Page Invoices, Receipts, and KYC ID DocumentsSize: 7B / 11B Vision

DocuVision-OCR-11B (Invoice, KYC & Financial Table Extraction SLM)

High-speed local vision-language model tuned for line-item invoice extraction, bank statement reconciliation, handwritten note digitization, and automated KYC document verification.

BASE ARCHITECTURE
Llama-3.2-11B-Vision-Instruct / Qwen2.5-VL-7B
VRAM REQUIREMENT
8 GB VRAM
QUANTIZATION FORMATS
Q4_K_M + mmproj-Q8_0 (6.2 GB), Q8_0 (9.8 GB)
PARAMETERS
7B / 11B Vision
QUICK DEPLOY COMMAND (OLLAMA / LLAMA.CPP):ollama run qwen2.5vl:7b

Local, Privacy-Preserving Document Intelligence

Instead of sending sensitive financial invoices or customer KYC identity documents to third-party cloud APIs, run DocuVision-OCR locally on an 8GB GPU and pipe structured line-item JSON straight into QuickBooks, Airtable, or PostgreSQL.