FreePublished on GitHub

donut

Use Donut for OCR-free document understanding, checkpoint inference, fine-tuning, evaluation, and SynthDoG synthetic document generation.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill

Works withHugging Face

See how to install

Files included (44)

SKILL.md4.7 KB
references/api-reference.md6.5 KB
references/repo-provenance.md3.6 KB
references/repo-routing-metadata.json489 B
references/troubleshooting.md5.2 KB
references/workflow-map.md5.2 KB
scripts/runtime_smoke.py6.5 KB
sub-skills/inference/references/demo-and-prediction.md6.3 KB
sub-skills/inference/references/troubleshooting.md4.5 KB
sub-skills/inference/scripts/launch_demo.py3.8 KB
sub-skills/inference/scripts/run_inference.py5.5 KB
sub-skills/inference/SKILL.md2.2 KB
sub-skills/synthdog/references/configs/config_en.yaml2.6 KB
sub-skills/synthdog/references/configs/config_ja.yaml2.6 KB
sub-skills/synthdog/references/configs/config_ko.yaml2.6 KB
sub-skills/synthdog/references/configs/config_zh.yaml2.6 KB
sub-skills/synthdog/references/configuration.md3.9 KB
sub-skills/synthdog/references/resource-layout.md2.0 KB
sub-skills/synthdog/references/troubleshooting.md3.3 KB
sub-skills/synthdog/references/workflows.md2.8 KB
sub-skills/synthdog/scripts/elements/__init__.py323 B
sub-skills/synthdog/scripts/elements/background.py608 B
sub-skills/synthdog/scripts/elements/content.py3.5 KB
sub-skills/synthdog/scripts/elements/document.py2.4 KB
sub-skills/synthdog/scripts/elements/paper.py391 B
sub-skills/synthdog/scripts/elements/textbox.py1.1 KB
sub-skills/synthdog/scripts/layouts/__init__.py169 B
sub-skills/synthdog/scripts/layouts/grid_stack.py2.4 KB
sub-skills/synthdog/scripts/layouts/grid.py2.3 KB
sub-skills/synthdog/scripts/render_config.py6.6 KB
sub-skills/synthdog/scripts/template.py5.0 KB
sub-skills/synthdog/SKILL.md2.5 KB
sub-skills/training/references/configs/train_cord.yaml940 B
sub-skills/training/references/configs/train_docvqa.yaml647 B
sub-skills/training/references/configs/train_rvlcdip.yaml659 B
sub-skills/training/references/configs/train_zhtrainticket.yaml578 B
sub-skills/training/references/configuration.md5.0 KB
sub-skills/training/references/troubleshooting.md4.1 KB
sub-skills/training/references/workflows.md5.0 KB
sub-skills/training/scripts/check_training_config.py12.4 KB
sub-skills/training/scripts/evaluate_dataset.py10.2 KB
sub-skills/training/scripts/lightning_module.py7.6 KB
sub-skills/training/scripts/train_donut.py6.6 KB
sub-skills/training/SKILL.md2.2 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

Donut

Use this repo skill when the task involves NAVER/ClovaAI Donut: OCR-free document understanding with a Swin encoder and BART-like decoder, plus SynthDoG synthetic document generation. This skill is self-contained; do not reopen the original checkout for routine Donut usage.

Route by task

User intent	Read or run
Load a Donut checkpoint, run one-image prediction, compare prompts, or start the Gradio demo	`sub-skills/inference/SKILL.md`
Fine-tune, validate dataset JSONL, evaluate a checkpoint, read metrics, or debug configs/checkpoints	`sub-skills/training/SKILL.md`
Generate synthetic document images, render SynthDoG configs, or adapt resources/fonts/corpora	`sub-skills/synthdog/SKILL.md`
Inspect public package APIs, signatures, token helpers, and object relationships	`references/api-reference.md`
Decide which bundled workflow helper or sub-skill owns a request	`references/workflow-map.md`
Diagnose install/import, Hugging Face, CUDA, Gradio, dataset, or SynthDoG dependency failures	`references/troubleshooting.md`
Check source baseline, commit, version, and refresh triggers	`references/repo-provenance.md`

Fast start

Install a Donut-capable Python environment. For inference only, pip install donut-python plus a compatible PyTorch build is the starting point. Training needs CUDA-capable PyTorch. SynthDoG generation also needs synthtiger and its image/text dependencies.

Smoke-check the environment without the source repo:

python scripts/runtime_smoke.py
python scripts/runtime_smoke.py --check all --require-cuda  # when training and SynthDoG are in scope

Route to a sub-skill before giving detailed commands. Each sub-skill owns its own scripts and troubleshooting notes.

Core facts to preserve

Package distribution: donut-python; import module: donut.
Public exports include DonutConfig, DonutModel, DonutDataset, JSONParseEvaluator, load_json, and save_json.
DonutModel.from_pretrained(...) loads local or Hugging Face checkpoints and requests the Donut repositories' official revision.
DonutModel.inference(...) returns a dictionary with a predictions list; parsed JSON is the default output.
Dataset rows use file_name plus a JSON-encoded ground_truth string containing either gt_parse or gt_parses.
Training is GPU-oriented in the original scripts; do not promise CPU training equivalence.
No model weights, datasets, or large SynthDoG assets are bundled in this skill. Users must provide checkpoints, images, datasets, and SynthDoG resources.

Bundled helpers

scripts/runtime_smoke.py: cross-workflow environment/API smoke check.
sub-skills/inference/scripts/run_inference.py: single-image inference and prompt comparison.
sub-skills/inference/scripts/launch_demo.py: Gradio demo launcher.
sub-skills/training/scripts/train_donut.py: bundled CUDA-oriented trainer entry point.
sub-skills/training/scripts/check_training_config.py: config and metadata validator.
sub-skills/training/scripts/evaluate_dataset.py: local/HF dataset evaluation helper and validation-only mode.
sub-skills/training/references/configs/: bundled CORD, DocVQA, RVL-CDIP, and TrainTicket config examples.
sub-skills/synthdog/scripts/render_config.py: render bundled SynthDoG placeholder configs for external resources.
sub-skills/synthdog/scripts/template.py: bundled SynthDoG template used by synthtiger.

Boundaries

Use Donut for OCR-free document-image understanding, classification, extraction, visual QA, and synthetic-document pretraining data generation. Do not use this skill for generic PDF parsing, OCR engines, RAG chunking, non-document computer vision models, or unrelated Transformers training unless Donut APIs, checkpoints, config files, errors, or data formats are central to the request.

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

More skills from this repository

3d-resnets-pytorch

Free

Routes 3D ResNets PyTorch video action-recognition workflows across training, inference, and data preparation.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill338Updated Sep 3, 2026

3ddfa

Free

Guide 3DDFA Python inference, geometry rendering, training/evaluation, and optional C++ ONNX workflows for 3D dense face alignment.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill338Updated Sep 3, 2026

3ddfa-v2

Free

Routes 3DDFA_V2 face-alignment setup, still-image demos, video tracking, and ONNX benchmarking workflows.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill338Updated Sep 3, 2026

ab3dmot

Free

Operate AB3DMOT 3D multi-object tracking workflows for KITTI and nuScenes data, tracking, evaluation, and visualization.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill338Updated Sep 3, 2026

accelerate

Free

Use Hugging Face Accelerate for PyTorch training-loop migration, distributed launch/configuration, DeepSpeed/FSDP/TPU backend setup, big-model inference/offload, checkpointing, tracking, and troubleshooting.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill338Updated Sep 3, 2026

acme

Free

Route Acme reinforcement-learning framework tasks across core loops, replay/data, JAX agents, and TensorFlow/Sonnet agents.

The English summary is being prepared. Showing the original description.

VectorSpaceLab/AREX-Skill338Updated Sep 3, 2026

See all skills by VectorSpaceLab

Report a problem with this skill

Install

Install with a command

npx skills add https://github.com/VectorSpaceLab/AREX-Skill --skill donut

Paste it into your terminal and run it (requires Node.js). You can choose which agents to install it for, such as Claude Code or Codex.

Ask your agent

Please install the following Agent Skill.
https://github.com/VectorSpaceLab/AREX-Skill/tree/main/skills/repositories/repo-skills/donut

First review SKILL.md and the bundled files and confirm there is nothing suspicious (such as sending data externally or reading credentials). Then place it in ~/.claude/skills/donut/ for Claude Code, or ~/.agents/skills/donut/ for Codex.

Send this text to Claude Code or Codex as is. It checks the contents before installing.

View source on GitHub

Repository: VectorSpaceLab/AREX-Skill
Stars: 338
Forks: 26
Category: Other
License: Apache-2.0
Original language: English
Last updated: Sep 3, 2026

A skill is a set of instructions and scripts for an agent. cccskills does not guarantee that skills are safe. Review the contents before you install.