lighton-ocr-2
lightonocr2-blind-spots
LightOnOCR-2-1B Blind Spots Dataset
Model Tested
lightonai/LightOnOCR-2-1B — a 1B parameter
end-to-end vision-language model for document OCR, released in 2026 under Apache 2.0.
How the Model Was Loaded
The model was loaded in Google Colab (free T4 GPU) using transformers installed from source:
import torch
from transformers import LightOnOcrForConditionalGeneration, LightOnOcrProcessor
device = "cuda"
dtype = torch.bfloat16
model =… See the full description on the dataset page: https://huggingface.co/datasets/Nonso-Analytics/lightonocr2-blind-spots.lighton-ocr2-test-v4
Document OCR using LightOnOCR-2-1B
This dataset contains OCR results from images in davanstrien/ufo-ColPali using LightOnOCR-2, a fast and compact 1B OCR model trained with RLVR.
Processing Details
Source Dataset: davanstrien/ufo-ColPali
Model: lightonai/LightOnOCR-2-1B
Number of Samples: 10
Processing Time: 2.8 min
Processing Date: 2026-01-29 17:49 UTC
Configuration
Image Column: image
Output Column: markdown
Dataset Split: train
Batch Size: 16
Target… See the full description on the dataset page: https://huggingface.co/datasets/davanstrien/lighton-ocr2-test-v4.lighton-ocr2-test-v1
Document OCR using LightOnOCR-2-1B
This dataset contains OCR results from images in davanstrien/ufo-ColPali using LightOnOCR-2, a fast and compact 1B OCR model trained with RLVR.
Processing Details
Source Dataset: davanstrien/ufo-ColPali
Model: lightonai/LightOnOCR-2-1B
Number of Samples: 10
Processing Time: 2.9 min
Processing Date: 2026-01-29 14:30 UTC
Configuration
Image Column: image
Output Column: markdown
Dataset Split: train
Batch Size: 16
Target… See the full description on the dataset page: https://huggingface.co/datasets/davanstrien/lighton-ocr2-test-v1.