Skip to content
Computer vision

Escriba-OCR-1B: Handwritten text recognition preserving errors

2 min readupdated

The model specializes in digitizing handwritten student essays in Brazilian Portuguese. Unlike universal OCR systems and VLMs, it does not correct the author's errors, instead transcribing them accurately line by line. Built on the LightOnOCR-3-1B architecture (combining a Pixtral encoder and a Qwen3 decoder), it contains about 1 billion parameters. Trained on the BRESSAY dataset, it preserves strikethroughs, corrections, and hyphenated line breaks using special markup. The model is useful for educators, researchers, and developers of educational tools analyzing real student work.

Key features

  • Handwriting specialization: Accurately decodes essay handwriting while preserving students' original spelling.
  • Preservation of edits: Transmits crossed-out words and superscript or subscript insertions using BRESSAY markup.
  • Compact size: Around 1 billion parameters allow running the model on both GPU and standard CPU via GGUF format.
  • High accuracy: Outperforms many large open models on the specialized dataset thanks to dedicated fine-tuning.

What you can do

  • Archive digitalization: Turn paper student papers into text for automated text corpus collection.
  • Linguistic research: Analyze the development of written language and typical student spelling errors.
  • Software integration: Use the model in local document processing pipelines using vLLM or llama.cpp.

Example

# pip install "vllm>=0.34" pillow
import base64, io
from PIL import Image
from vllm import LLM, SamplingParams

def to_data_url(path, long_side=1200):
    im = Image.open(path).convert("RGB")
    f = long_side / max(im.size)
    im = im.resize((round(im.width * f), round(im.height * f)), Image.LANCZOS)
    buf = io.BytesIO(); im.save(buf, format="PNG")
    return "data:image/png;base64," + base64.b64encode(buf.getvalue()).decode()

llm = LLM(model="tardellirs/Escriba-OCR-1B", max_model_len=8192, limit_mm_per_prompt={"image": 1})
params = SamplingParams(temperature=0.0, repetition_penalty=1.05, max_tokens=2048)

messages = [{"role": "user", "content": [{"type": "image_url", "image_url": {"url": to_data_url("redacao.jpg")}}]}]
print(llm.chat([messages], params)[0].outputs[0].text)
Read moreDetails
Benchmarks: CER (character error rate) is 2.06% on the BRESSAY test split, WER is 5.9%.
Hardware requirements: Efficient execution via vLLM on GPU or via llama.cpp on CPU (about 3.5 GB RAM for Q4_K_M quantization).
Limitations: CC BY-NC 4.0 license (non-commercial use only, inherited from training data).
Details: The model is not intended for evaluating essays, printed documents, or languages other than Portuguese.
Discuss Telegram 2 31

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.9 GB · ready files2.6 GB
  • FP8 / INT8recommendedweights 0.9 GB1.5 GB
  • 4-bitweights 0.5 GB1.1 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # tardellirs/Escriba-OCR-1B
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download tardellirs/Escriba-OCR-1B --local-dir ~/models/Escriba-OCR-1B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Escriba-OCR-1B
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of LightOnOCR-3-1B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 6All on Hugging Face
Fine-tunes 2All on Hugging Face
Merges 1All on Hugging Face

About the model

tardellirs/Escriba-OCR-1B
Author
tardellirs
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

312

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Computer vision

peshtur

peshtur