Document VLM FT: DocVQA + ChartQA + TableVQA + Türkçe Fatura/Dilekçe Dataset

1. Document VLM Dataset'leri#

Dataset	Size	Language	Notlar
DocVQA	50K	EN	document Q&A
ChartQA	21K	EN	chart understanding
TableVQA	23K	EN	table extraction
InfographicVQA	30K	EN	infographic Q&A
TR-Doc-Synthetic (cookbook)	10K	TR	senin oluşturduğun

# Synthetic Türkçe fatura üretimi (Playwright + HTML template)
from playwright.async_api import async_playwright

async def render_invoice(data):
    template = open("invoice_template.html").read()
    html = template.format(**data)

    async with async_playwright() as p:
        browser = await p.chromium.launch()
        page = await browser.new_page()
        await page.set_content(html)
        await page.screenshot(path=f"invoice_{data['id']}.png", full_page=True)
        await browser.close()

# 10K random fatura: ~30 dakika 4090 (CPU-bound)