bs-vision-train-v03 / README.md
shaban2024's picture
dizajn: banner, bedzevi, tabele
b8a6db6 verified
|
Raw History Blame Contribute Delete
2.2 kB
---
language:
- bs
license: cc-by-4.0
task_categories:
- image-text-to-text
- visual-question-answering
- document-question-answering
tags:
- bosnian
- ocr
- chart-qa
- synthetic
- bh-language-ai
---
![BH-Vision banner](https://raw.githubusercontent.com/BH-Language-AI/bh-vision/main/assets/banner-bh-vision.png)
[![License: CC-BY-4.0](https://img.shields.io/badge/License-CC_BY_4.0-lightgrey.svg)](https://creativecommons.org/licenses/by/4.0/)
[![Lang: bs](https://img.shields.io/badge/lang-bs-blue.svg)](https://huggingface.co/datasets?language=bs)
[![Uzoraka: 2179](https://img.shields.io/badge/uzoraka-2179-green.svg)](https://github.com/BH-Language-AI/bh-vision)
# 📦 bs-vision-train-v03 — sintetički bs trening set (2179 uzoraka, split po slici)
Vlastiti generisani sadržaj (100% čista licenca, CC-BY-4.0). Generator: [`scripts/make_synth_bs.py`](https://github.com/BH-Language-AI/bh-vision/blob/main/scripts/make_synth_bs.py) (seed 43, `OUT=data/synth3 N_DOC=800 N_CHART=400`).
| Split | Uzoraka | Slika |
|---|---|---|
| `data/synth3/train.jsonl` | 1962 | 1080 |
| `data/synth3/val.jsonl` | 217 | 120 |
| Slike ukupno (`data/synth3/img/`, jpg, ~32MB) | — | 1200 |
Split je **po slici** (0 preklapanja train/val) — v01/v02 su zastarjeli zbog curenja.
## Tipovi i pitanja
- 🧾 Računi: ukupan iznos, broj računa, broj stavki
- 🧾 Fakture-tabele: cijena artikla, broj stavki
- 📄 Obavještenja/potvrde/opomene/rješenja: naslov, ime građanina — sve sa čćdžšđ (30% lagana rotacija)
- 🧾 Izvodi: stanje računa
- 📊 Bar chartovi (max mjesec, vrijednost, razlika max-min) i line chartovi temperatura (min, vrijednost)
## Format i pravila
- Image-text-to-text chat JSONL: `messages[{role, content:[{type:image},{type:text}]}]` + `images` (putanje relativne na root frameworka)
- Font DejaVu Sans; chartovi vlastiti podaci (matplotlib)
- ⛔ NIKAD miješati sa held-outom [`bs-vision-ocr-heldout-v01`](https://huggingface.co/datasets/BH-Language-AI/bs-vision-ocr-heldout-v01) niti proxy evalom (XM3600-hr, postocr-sr)
- Trening: [`configs/smo*-qlora.yaml`](https://github.com/BH-Language-AI/bh-vision/tree/main/configs), start SmolVLM2-2.2B