Qwen3-VL-4B-Instruct-docling-ko-2510

Summary

VLM Model trained to be using with docling package's VLM Pipeline

  • Trained to convert given image to docling ResponseFormat.DOCTAGS format (ex. "...")
  • Training Data: 9K Pages of Korean FSC Press Release PDF Documents converted to doctags using docling=v2.55.0
  • Base Model: Qwen/Qwen3-VL-4B-Instruct + LoRA Adapter (r=16, alpha=32, dropout=0.05)

Usage

Refer to docling's VLM Pipeline documentation for detailed usage

Docling Converter with InlineVlm

  • provide artifacts_path for loading from local directory
from docling.datamodel.base_models import InputFormat
from docling.datamodel.accelerator_options import AcceleratorDevice
from docling.datamodel.pipeline_options import VlmPipelineOptions
from docling.datamodel.pipeline_options_vlm_model import (
    InferenceFramework,
    InlineVlmOptions,
    ResponseFormat,
    TransformersModelType
)
from docling.document_converter import DocumentConverter, PdfFormatOption
from docling.pipeline.vlm_pipeline import VlmPipeline

pipeline_options = VlmPipelineOptions(
    vlm_options=InlineVlmOptions(
        repo_id="id4thomas/Qwen3-VL-4B-Instruct-docling-ko-2510",
        prompt="Convert this page to docling.",
        response_format=ResponseFormat.DOCTAGS,
        inference_framework=InferenceFramework.TRANSFORMERS,
        transformers_model_type=TransformersModelType.AUTOMODEL_IMAGETEXTTOTEXT,
        supported_devices=[AcceleratorDevice.CUDA, AcceleratorDevice.CPU],
        scale=2.0,
        temperature=0.0,
    ),
)
converter = DocumentConverter(
    format_options={
        InputFormat.PDF: PdfFormatOption(
            pipeline_cls=VlmPipeline,
            pipeline_options=pipeline_options,
        ),
    }
)
doc = converter.convert(source=source).document

Conversion Examples

Conversion Examples using PDF files from allganize/RAG-Evaluation-Dataset-KO dataset

Examples from finance/[แ„‡แ…งแ†ฏแ„Žแ…ฅแ†ท] แ„Œแ…ตแ„‡แ…กแ†ผแ„‹แ…ณแ†ซแ„’แ…ขแ†ผแ„‹แ…ด แ„‰แ…ตแ„Œแ…ฎแ†ผแ„‹แ…ณแ†ซแ„’แ…ขแ†ผ แ„Œแ…ฅแ†ซแ„’แ…ชแ†ซแ„‰แ…ต แ„‹แ…ตแ†ซแ„€แ…กแ„‡แ…กแ†ผแ„‰แ…ตแ†จ แ„†แ…ตแ†พ แ„Œแ…ฅแ†ฏแ„Žแ…ก.pdf

Page 3

Example1
  • Black BBOX is drawn for each <loc_*> tag

VLM Result:

<doctag>
<section_header_level_1><loc_65><loc_38><loc_150><loc_48>โ…  .  ๊ฒ€ํ†  ๋ฐฐ๊ฒฝ</section_header_level_1>
<unordered_list><list_item><loc_60><loc_68><loc_447><loc_92>โ–ก ์ •๋ถ€๋Š” ์€ํ–‰๊ถŒ ๊ฒฝ์Ÿ์ด‰์ง„์„ ์œ„ํ•ด ์ง€๋ฐฉ์€ํ–‰์˜ ์‹œ์ค‘์€ํ–‰ ์ „ํ™˜์„ ํ—ˆ์šฉํ•˜๊ฒ ๋‹ค๊ณ  ๋ฐœํ‘œ * ('23.7.5 ์ผ )</list_item>
<list_item><loc_60><loc_100><loc_392><loc_107>'๊ธˆ์œต๋‹น๊ตญ, ์€ํ–‰์—…์— ๊ณต์ •ํ•˜๊ณ  ์‹คํšจ์„ฑ ์žˆ๋Š” ๊ฒฝ์Ÿ ๋„์ž…' (๋ณด๋„์ž๋ฃŒ)</list_item>
<list_item><loc_60><loc_120><loc_352><loc_127>โ€ป  ํ˜„์žฌ  ๋Œ€๊ตฌยท๊ฒฝ๋ถ๊ถŒ ์ง€๋ฐฉ์€ํ–‰์ธ ๋Œ€๊ตฌ์€ํ–‰์ด ์ „ํ™˜์˜์‚ฌ ํ‘œ๋ช…</list_item>
<list_item><loc_60><loc_140><loc_429><loc_151>โ–ก ํ˜„ํ–‰๋ฒ•๋ น์€ ' ์ง€๋ฐฉ้Š€ โ†’ ์‹œ์ค‘้Š€ ' ์ „ํ™˜์— ๊ด€ํ•œ ๋ช…์‹œ์  ๊ทœ์ • ์—†์Œ</list_item>
<list_item><loc_66><loc_163><loc_440><loc_186>๋‹ค๋งŒ , ์ง€๋ฐฉ์€ํ–‰์ด ์‹œ์ค‘์€ํ–‰ ์ธ๊ฐ€์š”๊ฑด์„ ๋ชจ๋‘ ๊ฐ–์ถ”๊ณ  ์ „ํ™˜ ์‹ ์ฒญ์‹œ ์ด๋ฅผ ํ—ˆ์šฉํ•˜์ง€ ์•Š์„ ๋ฒ•๋ น์ƒ ๊ทผ๊ฑฐ๋„ ์—†์œผ๋ฉฐ ,</list_item>
<list_item><loc_78><loc_194><loc_419><loc_205>-๊ฐ๋…์ •์ฑ…์  ์ธก๋ฉด์—์„œ๋„ ์ด๋ฅผ ๊ธˆ์ง€ํ•˜๋Š” ๊ฒƒ์€ ๊ณผ๋„ํ•œ ์ธก๋ฉด *</list_item>
<list_item><loc_60><loc_212><loc_444><loc_229>์‹ ๊ทœ์‚ฌ์—…์ž์™€  ๋™์ผํ•œ  ์š”๊ฑด์„  ๊ฐ–์ถ”์—ˆ์„  ๊ฒฝ์šฐ์—๋„  ๊ธฐ์กด์‚ฌ์—…์ž(์ง€๋ฐฉ์€ํ–‰)์— ๋Œ€ํ•ด์„œ๋งŒ ์‹œ์ค‘์€ํ–‰ ์ธ๊ฐ€๋ฅผ ๋ฐ›์ง€ ๋ชปํ•˜๋„๋ก ํ•˜๋Š” ๊ฒƒ์€ ๋ถˆํ•ฉ๋ฆฌํ•œ ์ธก๋ฉด</list_item>
<list_item><loc_66><loc_238><loc_440><loc_261>ํ•ฉ๋ณ‘ , ๊ณ„์•ฝ์ด์ „ (P&A) ์ด ์•„๋‹Œ ์€ํ–‰์˜ ์ข…๋ฅ˜ ์ „ํ™˜์€ ๊ณผ๊ฑฐ ์‚ฌ๋ก€๊ฐ€ ์—†์–ด ์ „ํ™˜๋ฐฉ์‹์— ๊ด€ํ•œ ๋ฒ•์ ๊ทผ๊ฑฐ ยท ์ ˆ์ฐจ ๋“ฑ ๊ฒ€ํ†  ํ•„์š”</list_item>
<list_item><loc_60><loc_272><loc_438><loc_279>ํ•ฉ๋ณ‘:  ํ•˜๋‚˜์€ํ–‰-์™ธํ™˜์€ํ–‰('15) / ๊ณ„์•ฝ์ด์ „: ๊ฒฝ๊ธฐยท์ถฉ์ฒญยท๋Œ€๋™ยท๋™๋‚จยท๋™ํ™”์€ํ–‰('98)</list_item>
<list_item><loc_60><loc_292><loc_447><loc_316>โ–ก ํ•œํŽธ , ์ง€๋ฐฉ์€ํ–‰์ด ์ž์ฒด์ ์œผ๋กœ ์ •๊ด€ ๋ณ€๊ฒฝ์„ ํ†ตํ•ด ์‹œ์ค‘์€ํ–‰์œผ๋กœ ์ „ํ™˜์ด ๊ฐ€๋Šฅํ•˜๋‹ค๋Š” ์ผ๋ถ€ ์˜๊ฒฌ์ด ์žˆ์œผ๋‚˜ , ์ด๋Š” ๋ถ€์ ์ ˆํ•œ ์ธก๋ฉด</list_item>
<list_item><loc_66><loc_328><loc_440><loc_351>์ง€๋ฐฉ์€ํ–‰ ์˜์—…๊ตฌ์—ญ์€ ์ •๊ด€์—์„œ ํŠน์ •์ง€์—ญ์œผ๋กœ ์ œํ•œํ•˜๊ณ  ์žˆ๋Š”๋ฐ , ์ด๋ฅผ ์ „๊ตญ์œผ๋กœ ๋ณ€๊ฒฝ์‹œ ์ „ํ™˜์ด ๊ฐ€๋Šฅํ•˜๋‹ค๋Š” ํ•ด์„</list_item>
<list_item><loc_66><loc_364><loc_441><loc_387>ใ…‡ ํ•˜์ง€๋งŒ , ์€ํ–‰ ์ข…๋ฅ˜์˜ ์ „ํ™˜์€ ๊ธˆ์œต๊ฐ๋…์ •์ฑ…์˜ ์ค‘์š”์‚ฌํ•ญ์œผ๋กœ ์‚ฌ์ „ ์Šน์ธ์ ˆ์ฐจ ์—†์ด ์ •๊ด€ ๋ณ€๊ฒฝ ( โ˜ž ์‚ฌํ›„๋ณด๊ณ  ) ๋งŒ์œผ๋กœ ํ—ˆ์šฉํ•˜๋Š” ๊ฒƒ์€ ๊ณค๋ž€</list_item>
<list_item><loc_66><loc_408><loc_440><loc_432>โžก ์œ„ ์‚ฌํ•ญ์„ ์ข…ํ•ฉ์ ์œผ๋กœ ๊ณ ๋ คํ•˜์—ฌ ็พ ๋ฒ•๋ น์ฒด๊ณ„ ไธ‹ ์—์„œ ์ง€๋ฐฉ์€ํ–‰์˜ ์‹œ์ค‘์€ํ–‰ ์ „ํ™˜ ์ธ๊ฐ€์— ๊ด€ํ•œ ์ฃผ์š”์Ÿ์  ๋ฐ ๊ตฌ์ฒด์  ๋ฐฉ์•ˆ ๊ฒ€ํ† </list_item>
</unordered_list>
<page_footer><loc_240><loc_477><loc_260><loc_483>-  1  -</page_footer>
</doctag>

Page 7 (Table)

Example2

VLM Result:

<doctag>
<section_header_level_1><loc_65><loc_38><loc_106><loc_48>์ฐธ๊ณ  2</section_header_level_1>
<section_header_level_1><loc_117><loc_38><loc_283><loc_48>์€ํ–‰์—… ์ธ๊ฐ€ ์„ธ๋ถ€์‹ฌ์‚ฌ์š”๊ฑด</section_header_level_1>
<otsl><loc_63><loc_63><loc_436><loc_447><ecel><ched>์„ธ ๋ถ€ ์‹ฌ ์‚ฌ ์š” ๊ฑด<ched>ํ™•์ธ์„œ๋ฅ˜<nl><fcel>์ž๋ณธ๊ธˆ ์š”๊ฑด<fcel>ใ…‡ ์ตœ์ € ์ž๋ณธ๊ธˆ ์š”๊ฑด์„ ์ถฉ์กฑํ•  ๊ฒƒ ใ…‡ ์ž๊ธˆ์กฐ๋‹ฌ๋ฐฉ์•ˆ์ด ์ ์ •ํ•  ๊ฒƒ<fcel>- ์ž๋ณธ๊ธˆ ๋‚ฉ์ž… ํ™•์•ฝ์„œ ๋“ฑ<nl><fcel>๋Œ€์ฃผ์ฃผ ์š”๊ฑด<fcel>ใ…‡ ๋ถ€์‹ค๊ธˆ์œต๊ธฐ๊ด€ ๊ด€๋ จ ์ฑ…์ž„์ด ์—†์„ ๊ฒƒ ใ…‡ ์ฃผ์ฃผ๊ตฌ์„ฑ๊ณ„ํš์ด ์€ํ–‰๋ฒ•์ƒ ์†Œ์œ ๊ทœ์ œ์— ์ ํ•ฉํ•  ๊ฒƒ<fcel>- ๋น„๊ธˆ์œต์ฃผ๋ ฅ์ž๊ฐ€ ์•„๋‹˜์„ ์ฆ๋ช… ํ•˜๋Š” ์„œ๋ฅ˜ ๋“ฑ<nl><fcel>์‚ฌ์—…๊ณ„ํš ํƒ€๋‹น์„ฑ ์š”๊ฑด<fcel>ใ…‡ ๊ฒฝ์˜์ „๋žต ๋ฐ ์ˆ˜์ต์ „๋ง์ด ์ ์ •ํ•  ๊ฒƒ ใ…‡ ๊ฒฝ์˜์ง€๋„๊ธฐ์ค€ ์ถฉ์กฑ์ด ๊ฐ€๋Šฅํ•  ๊ฒƒ ใ…‡ ์ด์‚ฌํšŒ ๋ฐ ๊ฒฝ์˜์ง€๋ฐฐ๊ตฌ์กฐ๊ฐ€ ์ ์ •ํ•  ๊ฒƒ ใ…‡ ๋‚ด๋ถ€ํ†ต์ œ, ์ค€๋ฒ•๊ฐ์‹œ ๋ฐ ๋ฆฌ์Šคํฌ ๊ด€๋ฆฌ ์ฒด๊ณ„๊ฐ€ ์ ์ •ํ•  ๊ฒƒ ใ…‡ ์˜์—…๋‚ด์šฉ ๋ฐ ๋ฐฉ๋ฒ•์ด ๋ฒ•๋ น ๋ฐ ๊ฑด์ „ํ•œ ๊ธˆ์œต๊ฑฐ๋ž˜์งˆ์„œ์— ๋ถ€ํ•ฉํ•  ๊ฒƒ<fcel>- ์‹ ์ฒญ์„œ์ƒ ์‚ฌ์—…๊ณ„ํš์„œ๋“ฑ<nl><fcel>์ž„์› ์š”๊ฑด<fcel>ใ…‡ ๋ฐœ๊ธฐ์ธ ๋ฐ ์ž„์›์ด ์€ํ–‰๋ฒ•์ƒ ์ž„์›์ž๊ฒฉ ์š”๊ฑด์— ๋ถ€ํ•ฉํ•  ๊ฒƒ<fcel>- ๊ฒฝ๋ ฅ์ฆ๋ช…์„œ, ์ž๊ฒฉ์ฆ ๋“ฑ - ์‹ ์›์กฐํšŒ ๋ฐ ๊ด€๋ จ๋ถ€์„œ ์‚ฌ์‹ค ์กฐํšŒ ํšŒ๋ณด์„œ<nl><fcel>์ธ๋ ฅยท์˜์—…์‹œ์„คยท ์ „์‚ฐ์„ค๋น„ ์š”๊ฑด<fcel>ใ…‡ ์ธ๊ฐ€์‹ ์ฒญ์—…๋ฌด๋ฅผ ์˜์œ„ํ•˜๊ธฐ ์œ„ํ•œ ์ธ๋ ฅ (์ „๋ฌธ์ธ๋ ฅ ํฌํ•จ) ํ™•๋ณด๊ณ„ํš์ด ์ ์ •ํ•  ๊ฒƒ ใ…‡ ์—…๋ฌด๋ฒ”์œ„ ๋ฐ ๊ทœ๋ชจ์— ๋ถ€ํ•ฉํ•˜๋Š” ์˜์—… ์‹œ์„ค ๋ฐ ์ดํ•ด์ƒ์ถฉ๋ฐฉ์ง€์ฒด๊ณ„๋ฅผ ๊ฐ–์ถœ ๊ฒƒ ใ…‡ ์€ํ–‰์—… ์˜์œ„๋ฅผ ์œ„ํ•œ ์ ์ •ํ•œ ์ „์‚ฐ์„ค๋น„๋ฅผ ๊ฐ–์ถœ ๊ฒƒ<fcel>- ์‹ ์ฒญ์„œ์ƒ ์‚ฌ์—…๊ณ„ํš์„œ๋“ฑ<nl></otsl>
<page_footer><loc_239><loc_477><loc_260><loc_483>-  5  -</page_footer>
</doctag>

Table Converted to HTML:

<table>
<tbody>
    <tr>
        <td></td>
        <td>์„ธ ๋ถ€ ์‹ฌ ์‚ฌ ์š” ๊ฑด</td>
        <td>ํ™•์ธ์„œ๋ฅ˜</td>
    </tr>
    <tr>
        <td>์ž๋ณธ๊ธˆ ์š”๊ฑด</td>
        <td>ใ…‡ ์ตœ์ € ์ž๋ณธ๊ธˆ ์š”๊ฑด์„ ์ถฉ์กฑํ•  ๊ฒƒ ใ…‡ ์ž๊ธˆ์กฐ๋‹ฌ๋ฐฉ์•ˆ์ด ์ ์ •ํ•  ๊ฒƒ</td>
        <td>- ์ž๋ณธ๊ธˆ ๋‚ฉ์ž… ํ™•์•ฝ์„œ ๋“ฑ</td>
    </tr>
    <tr>
        <td>๋Œ€์ฃผ์ฃผ ์š”๊ฑด</td>
        <td>ใ…‡ ๋ถ€์‹ค๊ธˆ์œต๊ธฐ๊ด€ ๊ด€๋ จ ์ฑ…์ž„์ด ์—†์„ ๊ฒƒ ใ…‡ ์ฃผ์ฃผ๊ตฌ์„ฑ๊ณ„ํš์ด ์€ํ–‰๋ฒ•์ƒ ์†Œ์œ ๊ทœ์ œ์— ์ ํ•ฉํ•  ๊ฒƒ</td>
        <td>- ๋น„๊ธˆ์œต์ฃผ๋ ฅ์ž๊ฐ€ ์•„๋‹˜์„ ์ฆ๋ช… ํ•˜๋Š” ์„œ๋ฅ˜ ๋“ฑ</td>
    </tr>
    <tr>
        <td>์‚ฌ์—…๊ณ„ํš ํƒ€๋‹น์„ฑ ์š”๊ฑด</td>
        <td>ใ…‡ ๊ฒฝ์˜์ „๋žต ๋ฐ ์ˆ˜์ต์ „๋ง์ด ์ ์ •ํ•  ๊ฒƒ ใ…‡ ๊ฒฝ์˜์ง€๋„๊ธฐ์ค€ ์ถฉ์กฑ์ด ๊ฐ€๋Šฅํ•  ๊ฒƒ ใ…‡ ์ด์‚ฌํšŒ ๋ฐ ๊ฒฝ์˜์ง€๋ฐฐ๊ตฌ์กฐ๊ฐ€ ์ ์ •ํ•  ๊ฒƒ ใ…‡ ๋‚ด๋ถ€ํ†ต์ œ, ์ค€๋ฒ•๊ฐ์‹œ ๋ฐ ๋ฆฌ์Šคํฌ ๊ด€๋ฆฌ ์ฒด๊ณ„๊ฐ€ ์ ์ •ํ•  ๊ฒƒ ใ…‡ ์˜์—…๋‚ด์šฉ ๋ฐ ๋ฐฉ๋ฒ•์ด ๋ฒ•๋ น ๋ฐ ๊ฑด์ „ํ•œ ๊ธˆ์œต๊ฑฐ๋ž˜์งˆ์„œ์— ๋ถ€ํ•ฉํ•  ๊ฒƒ</td>
        <td>- ์‹ ์ฒญ์„œ์ƒ ์‚ฌ์—…๊ณ„ํš์„œ๋“ฑ</td>
    </tr>
    <tr>
        <td>์ž„์› ์š”๊ฑด</td>
        <td>ใ…‡ ๋ฐœ๊ธฐ์ธ ๋ฐ ์ž„์›์ด ์€ํ–‰๋ฒ•์ƒ ์ž„์›์ž๊ฒฉ ์š”๊ฑด์— ๋ถ€ํ•ฉํ•  ๊ฒƒ</td>
        <td>- ๊ฒฝ๋ ฅ์ฆ๋ช…์„œ, ์ž๊ฒฉ์ฆ ๋“ฑ - ์‹ ์›์กฐํšŒ ๋ฐ ๊ด€๋ จ๋ถ€์„œ ์‚ฌ์‹ค ์กฐํšŒ ํšŒ๋ณด์„œ</td>
    </tr>
    <tr>
        <td>์ธ๋ ฅยท์˜์—…์‹œ์„คยท ์ „์‚ฐ์„ค๋น„ ์š”๊ฑด</td>
        <td>ใ…‡ ์ธ๊ฐ€์‹ ์ฒญ์—…๋ฌด๋ฅผ ์˜์œ„ํ•˜๊ธฐ ์œ„ํ•œ ์ธ๋ ฅ (์ „๋ฌธ์ธ๋ ฅ ํฌํ•จ) ํ™•๋ณด๊ณ„ํš์ด ์ ์ •ํ•  ๊ฒƒ ใ…‡ ์—…๋ฌด๋ฒ”์œ„ ๋ฐ ๊ทœ๋ชจ์— ๋ถ€ํ•ฉํ•˜๋Š” ์˜์—… ์‹œ์„ค ๋ฐ ์ดํ•ด์ƒ์ถฉ๋ฐฉ์ง€์ฒด๊ณ„๋ฅผ ๊ฐ–์ถœ ๊ฒƒ ใ…‡ ์€ํ–‰์—… ์˜์œ„๋ฅผ ์œ„ํ•œ ์ ์ •ํ•œ ์ „์‚ฐ์„ค๋น„๋ฅผ ๊ฐ–์ถœ ๊ฒƒ</td>
        <td>- ์‹ ์ฒญ์„œ์ƒ ์‚ฌ์—…๊ณ„ํš์„œ๋“ฑ</td>
    </tr>
</tbody>
</table>
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for id4thomas/Qwen3-VL-4B-Instruct-docling-ko-2510

Finetuned
(29)
this model

Collection including id4thomas/Qwen3-VL-4B-Instruct-docling-ko-2510