Nemotron Parse
Registry ids: nemotron-parse (hosted), nemotron-parse-local (Docker NIM).
Model id: nvidia/nemotron-parse. Image in → markdown (and optional bounding boxes) out. No text-only input.
Hosted
export NVIDIA_API_KEY=...
opentryon parse --model nemotron-parse --image invoice.png
Local NIM
docker run ... nvcr.io/nim/nvidia/nemotron-parse:1.5.0
export NEMOTRON_PARSE_BASE_URL=http://127.0.0.1:8000/v1
opentryon parse --model nemotron-parse-local --image scan.jpg --tool markdown_bbox
Tools
--tool | Output |
|---|---|
markdown_bbox | Markdown + normalized boxes (default) |
markdown_no_bbox | Markdown only |
detection_only | Layout boxes without transcription |
Response JSON includes markdown, blocks, and raw tool payload.
Open-weight alternative on the same service: LightOnOCR-3 (lightonocr-3-local, lightonocr-3-vllm; also reads PDFs).