Skip to content

PAIDEIA v0.9.25 — vision_ocr grows a --model switch

Choose a tag to compare

View all tags
@TaewoooPark TaewoooPark released this 04 Jul 20:04
· 12 commits to main since this release
v0.9.25
3eb98ee

feat(vision-ocr): --model= + OLLAMA_VISION_MODEL

The ollama OCR tier's vision model was hardcoded (qwen3-vl:8b). Hosts that embed these scripts — OPTIMETA-OS routes codex/ollama-engine OCR through vision_ocr.py — can now aim it at ollama-cloud vision models (e.g. qwen3-vl:235b-cloud) per call or per environment:

python3 vision_ocr.py --engine=ollama --model=qwen3-vl:235b-cloud in.pdf out.md
OLLAMA_VISION_MODEL=qwen3-vl:235b-cloud python3 vision_ocr.py --engine=ollama …

Precedence: flag > env > local default. The default stays the local 8B — offline, no account needed. 55 tests green.

Assets 2
Loading