SonYiHF/Verus-OCR
0
Verus-OCR
Verus-OCR is a scratch OCR model project by 8F-ai.
Status: training code is staged, but the first H200 Hugging Face Jobs launch was blocked by insufficient prepaid Jobs credits on the active account.
Requested first run:
- Hardware: Hugging Face Jobs
h200 - Duration: 1 hour timeout, 52 minute training budget
- Dataset:
Teklia/IAM-line, a real handwritten OCR dataset - Architecture:
VerusOCRForCasualLM - Target size: about 500M parameters
max_position_embeddings: 8190
Run after adding Jobs credits:
hf jobs uv run --flavor h200 --timeout 1h --secrets HF_TOKEN --label project=verus-ocr --label model=Verus-OCR --detach train_verus_ocr.py --repo-id SonYiHF/Verus-OCR --run-minutes 52 --max-steps 1600 --dataset-name Teklia/IAM-line --max-position-embeddings 8190Use --repo-id 8F-ai/Verus-OCR once the active token has write access to the 8F-ai namespace.
