experiments/exp_02.yaml

name: docvision_full_pipeline
task: document_classification
model: nvidia/nemotron-nano-12b-v2-vl:free

ocr_engine: llamaparse
use_visual_cues: true
logo_detection_model: ellabettison/Logo-Detection-finetune

max_pages: 1
max_logos_per_page: 4
image_resize: [1024, 1024]

temperature: 0.1
seed: 42

description: >
  Full DocVision pipeline experiment combining OCR, Vision LLM
  reasoning, and visual cue detection. Logos and seals extracted
  from documents are used to support document classification
  and improve robustness on visually distinctive documents.