cstr/lightonocr-GGUF

LightOnOCR-2-1B — GGUF

GGUF conversion of lightonai/LightOnOCR-2-1B for CrispEmbed.

OCR Arena #2 (ELO 1697, 72% win rate). End-to-end document OCR — converts scans, PDFs, and photos to clean markdown.

Architecture

Component	Details
Vision	Pixtral ViT (24L, 1024d, 2D RoPE, SiLU FFN)
Projection	2×2 spatial merge + 2-layer MLP + RMSNorm
Decoder	Qwen3 (28L, 1024d, GQA 16/8, QK norm, SwiGLU)
Total	1006M params
License	Apache 2.0

Available Formats

File	Format	Size
lightonocr-1b-f16.gguf	F16	2218 MB
lightonocr-1b-q8_0.gguf	Q8_0	1025 MB
lightonocr-1b-q4_k.gguf	Q4_K	622 MB

Usage

crispembed -m lightonocr-1b-q4_k.gguf --ocr document.png

Auto-detected from GGUF architecture metadata (lightonocr).

Provenance and EU AI Act Art. 53 note

• Upstream model: lightonai/LightOnOCR-2-1B — published by lightonai.
• Upstream licence: apache-2.0. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.