{
"cells": [
{
"cell_type": "markdown",
"id": "922d396f",
"metadata": {},
"source": [
"# Table annotations"
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "50437c89",
"metadata": {},
"outputs": [],
"source": [
"from docling_core.types.doc.document import DoclingDocument\n",
"\n",
"file_path = \"2408.09869v3.json\"\n",
"pages = {5} # pages to serialize (for output brevity)\n",
"\n",
"doc = DoclingDocument.load_from_json(file_path)"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "d35192ea",
"metadata": {},
"outputs": [],
"source": [
"from typing import Optional\n",
"from rich.console import Console\n",
"from rich.panel import Panel\n",
"\n",
"\n",
"def print_excerpt(\n",
" txt: str,\n",
" *,\n",
" limit: int = 2000,\n",
" title: Optional[str] = None,\n",
" min_width: int = 80,\n",
" table_end: str = \"--|\",\n",
"):\n",
" excerpt = txt[:limit]\n",
" width = max(\n",
" max([ln.rfind(table_end) for ln in excerpt.splitlines()]) + len(table_end) + 4,\n",
" min_width,\n",
" )\n",
" console = Console(width=width)\n",
" console.print(Panel(f\"{excerpt}{'...' if len(txt) > limit else ''}\", title=title))"
]
},
{
"cell_type": "markdown",
"id": "a51271ac",
"metadata": {},
"source": [
"## Adding a table annotation"
]
},
{
"cell_type": "markdown",
"id": "557791de",
"metadata": {},
"source": [
"Below we add a demo table annotation, picking the first table for illustrative purposes.\n",
"\n",
"Note that `TableMiscData` allows any dict data within the `content` field.\n"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "add64711",
"metadata": {},
"outputs": [],
"source": [
"from docling_core.types.doc.document import DescriptionAnnotation, MiscAnnotation\n",
"\n",
"assert doc.tables, \"No table available in this document\"\n",
"table = doc.tables[0]\n",
"\n",
"table.add_annotation(\n",
" annotation=DescriptionAnnotation(\n",
" text=\"A typical Docling setup runtime characterization.\",\n",
" provenance=\"model-foo\",\n",
" ),\n",
")\n",
"\n",
"table.add_annotation(\n",
" annotation=MiscAnnotation(\n",
" content={\n",
" \"type\": \"performance data\",\n",
" \"sentiment\": 0.85,\n",
" # ...\n",
" },\n",
" ),\n",
")"
]
},
{
"cell_type": "markdown",
"id": "81408ae6",
"metadata": {},
"source": [
"## Default serialization"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "b1be8540",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"╭────────────────────────────────────────────────────────────────────────────────── pages={5} ───────────────────────────────────────────────────────────────────────────────────╮\n",
"│ torch runtimes backing the Docling pipeline. We will deliver updates on this topic at in a future version of this report. │\n",
"│ │\n",
"│ Table 1: Runtime characteristics of Docling with the standard model pipeline and settings, on our test dataset of 225 pages, on two different systems. OCR is disabled. We │\n",
"│ show the time-to-solution (TTS), computed throughput in pages per second, and the peak memory used (resident set size) for both the Docling-native PDF backend and for the │\n",
"│ pypdfium backend, using 4 and 16 threads. │\n",
"│ │\n",
"│ A typical Docling setup runtime characterization. │\n",
"│ │\n",
"│ | CPU | Thread budget | native backend | native backend | native backend | pypdfium backend | pypdfium backend | pypdfium backend | │\n",
"│ |----------------------------------|-----------------|------------------|------------------|------------------|--------------------|--------------------|--------------------| │\n",
"│ | | | TTS | Pages/s | Mem | TTS | Pages/s | Mem | │\n",
"│ | Apple M3 Max | 4 | 177 s 167 s | 1.27 1.34 | 6.20 GB | 103 s 92 s | 2.18 2.45 | 2.56 GB | │\n",
"│ | (16 cores) Intel(R) Xeon E5-2690 | 16 4 16 | 375 s 244 s | 0.60 0.92 | 6.16 GB | 239 s 143 s | 0.94 1.57 | 2.42 GB | │\n",
"│ │\n",
"│ ## 5 Applications │\n",
"│ │\n",
"│ Thanks to the high-quality, richly structured document conversion achieved by Docling, its output qualifies for numerous downstream applications. For example, Docling can │\n",
"│ provide a base for detailed enterprise document search, passage retrieval or classification use-cases, or support knowledge extraction pipelines, allowing specific treatment │\n",
"│ of different structures in the document, such as tables, figures, section structure or references. For popular generative AI application patterns, such as retrieval-augmented │\n",
"│ generation (RAG), we provi... │\n",
"╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯\n",
"\n"
],
"text/plain": [
"╭────────────────────────────────────────────────────────────────────────────────── pages={5} ───────────────────────────────────────────────────────────────────────────────────╮\n",
"│ torch runtimes backing the Docling pipeline. We will deliver updates on this topic at in a future version of this report. │\n",
"│ │\n",
"│ Table 1: Runtime characteristics of Docling with the standard model pipeline and settings, on our test dataset of 225 pages, on two different systems. OCR is disabled. We │\n",
"│ show the time-to-solution (TTS), computed throughput in pages per second, and the peak memory used (resident set size) for both the Docling-native PDF backend and for the │\n",
"│ pypdfium backend, using 4 and 16 threads. │\n",
"│ │\n",
"│ A typical Docling setup runtime characterization. │\n",
"│ │\n",
"│ | CPU | Thread budget | native backend | native backend | native backend | pypdfium backend | pypdfium backend | pypdfium backend | │\n",
"│ |----------------------------------|-----------------|------------------|------------------|------------------|--------------------|--------------------|--------------------| │\n",
"│ | | | TTS | Pages/s | Mem | TTS | Pages/s | Mem | │\n",
"│ | Apple M3 Max | 4 | 177 s 167 s | 1.27 1.34 | 6.20 GB | 103 s 92 s | 2.18 2.45 | 2.56 GB | │\n",
"│ | (16 cores) Intel(R) Xeon E5-2690 | 16 4 16 | 375 s 244 s | 0.60 0.92 | 6.16 GB | 239 s 143 s | 0.94 1.57 | 2.42 GB | │\n",
"│ │\n",
"│ ## 5 Applications │\n",
"│ │\n",
"│ Thanks to the high-quality, richly structured document conversion achieved by Docling, its output qualifies for numerous downstream applications. For example, Docling can │\n",
"│ provide a base for detailed enterprise document search, passage retrieval or classification use-cases, or support knowledge extraction pipelines, allowing specific treatment │\n",
"│ of different structures in the document, such as tables, figures, section structure or references. For popular generative AI application patterns, such as retrieval-augmented │\n",
"│ generation (RAG), we provi... │\n",
"╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"from docling_core.transforms.serializer.markdown import (\n",
" MarkdownDocSerializer,\n",
" MarkdownParams,\n",
")\n",
"\n",
"ser = MarkdownDocSerializer(\n",
" doc=doc,\n",
" params=MarkdownParams(\n",
" pages=pages,\n",
" ),\n",
")\n",
"ser_out = ser.serialize()\n",
"ser_txt = ser_out.text\n",
"\n",
"print_excerpt(ser_txt, title=f\"{pages=}\")"
]
},
{
"cell_type": "markdown",
"id": "50b513c1",
"metadata": {},
"source": [
"## Custom serialization"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "add5b785",
"metadata": {},
"outputs": [],
"source": [
"from typing import Any\n",
"\n",
"from docling_core.transforms.serializer.base import SerializationResult\n",
"from docling_core.transforms.serializer.common import create_ser_result\n",
"from docling_core.transforms.serializer.markdown import MarkdownAnnotationSerializer\n",
"from docling_core.types.doc.document import MiscAnnotation, DocItem\n",
"\n",
"\n",
"class CustomAnnotationSerializer(MarkdownAnnotationSerializer):\n",
" def serialize(\n",
" self,\n",
" *,\n",
" item: DocItem,\n",
" doc: DoclingDocument,\n",
" **kwargs: Any,\n",
" ) -> SerializationResult:\n",
" text_parts: list[str] = []\n",
"\n",
" # reusing result from parent serializer:\n",
" parent_res = super().serialize(\n",
" item=item,\n",
" doc=doc,\n",
" **kwargs,\n",
" )\n",
" text_parts.append(parent_res.text)\n",
"\n",
" # custom serialization logic (appending misc annotation result):\n",
" for ann in item.get_annotations():\n",
" if isinstance(ann, MiscAnnotation):\n",
" out_txt = \"\".join([f\"- {k}: {ann.content[k]}\\n\" for k in ann.content])\n",
" text_parts.append(out_txt)\n",
" text_res = \"\\n\\n\".join(text_parts)\n",
" return create_ser_result(text=text_res, span_source=item)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "e1107ddb",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"╭────────────────────────────────────────────────────────────────────────────────── pages={5} ───────────────────────────────────────────────────────────────────────────────────╮\n",
"│ torch runtimes backing the Docling pipeline. We will deliver updates on this topic at in a future version of this report. │\n",
"│ │\n",
"│ Table 1: Runtime characteristics of Docling with the standard model pipeline and settings, on our test dataset of 225 pages, on two different systems. OCR is disabled. We │\n",
"│ show the time-to-solution (TTS), computed throughput in pages per second, and the peak memory used (resident set size) for both the Docling-native PDF backend and for the │\n",
"│ pypdfium backend, using 4 and 16 threads. │\n",
"│ │\n",
"│ A typical Docling setup runtime characterization. │\n",
"│ │\n",
"│ - type: performance data │\n",
"│ - sentiment: 0.85 │\n",
"│ │\n",
"│ │\n",
"│ | CPU | Thread budget | native backend | native backend | native backend | pypdfium backend | pypdfium backend | pypdfium backend | │\n",
"│ |----------------------------------|-----------------|------------------|------------------|------------------|--------------------|--------------------|--------------------| │\n",
"│ | | | TTS | Pages/s | Mem | TTS | Pages/s | Mem | │\n",
"│ | Apple M3 Max | 4 | 177 s 167 s | 1.27 1.34 | 6.20 GB | 103 s 92 s | 2.18 2.45 | 2.56 GB | │\n",
"│ | (16 cores) Intel(R) Xeon E5-2690 | 16 4 16 | 375 s 244 s | 0.60 0.92 | 6.16 GB | 239 s 143 s | 0.94 1.57 | 2.42 GB | │\n",
"│ │\n",
"│ ## 5 Applications │\n",
"│ │\n",
"│ Thanks to the high-quality, richly structured document conversion achieved by Docling, its output qualifies for numerous downstream applications. For example, Docling can │\n",
"│ provide a base for detailed enterprise document search, passage retrieval or classification use-cases, or support knowledge extraction pipelines, allowing specific treatment │\n",
"│ of different structures in the document, such as tables, figures, section structure or references. For popular generative AI application patterns, such as r... │\n",
"╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯\n",
"\n"
],
"text/plain": [
"╭────────────────────────────────────────────────────────────────────────────────── pages={5} ───────────────────────────────────────────────────────────────────────────────────╮\n",
"│ torch runtimes backing the Docling pipeline. We will deliver updates on this topic at in a future version of this report. │\n",
"│ │\n",
"│ Table 1: Runtime characteristics of Docling with the standard model pipeline and settings, on our test dataset of 225 pages, on two different systems. OCR is disabled. We │\n",
"│ show the time-to-solution (TTS), computed throughput in pages per second, and the peak memory used (resident set size) for both the Docling-native PDF backend and for the │\n",
"│ pypdfium backend, using 4 and 16 threads. │\n",
"│ │\n",
"│ A typical Docling setup runtime characterization. │\n",
"│ │\n",
"│ - type: performance data │\n",
"│ - sentiment: 0.85 │\n",
"│ │\n",
"│ │\n",
"│ | CPU | Thread budget | native backend | native backend | native backend | pypdfium backend | pypdfium backend | pypdfium backend | │\n",
"│ |----------------------------------|-----------------|------------------|------------------|------------------|--------------------|--------------------|--------------------| │\n",
"│ | | | TTS | Pages/s | Mem | TTS | Pages/s | Mem | │\n",
"│ | Apple M3 Max | 4 | 177 s 167 s | 1.27 1.34 | 6.20 GB | 103 s 92 s | 2.18 2.45 | 2.56 GB | │\n",
"│ | (16 cores) Intel(R) Xeon E5-2690 | 16 4 16 | 375 s 244 s | 0.60 0.92 | 6.16 GB | 239 s 143 s | 0.94 1.57 | 2.42 GB | │\n",
"│ │\n",
"│ ## 5 Applications │\n",
"│ │\n",
"│ Thanks to the high-quality, richly structured document conversion achieved by Docling, its output qualifies for numerous downstream applications. For example, Docling can │\n",
"│ provide a base for detailed enterprise document search, passage retrieval or classification use-cases, or support knowledge extraction pipelines, allowing specific treatment │\n",
"│ of different structures in the document, such as tables, figures, section structure or references. For popular generative AI application patterns, such as r... │\n",
"╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯\n"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"ser = MarkdownDocSerializer(\n",
" doc=doc,\n",
" annotation_serializer=CustomAnnotationSerializer(),\n",
" params=MarkdownParams(\n",
" pages=pages,\n",
" ),\n",
")\n",
"ser_out = ser.serialize()\n",
"ser_txt = ser_out.text\n",
"\n",
"print_excerpt(ser_txt, title=f\"{pages=}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "fb350716",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": ".venv",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.12.4"
}
},
"nbformat": 4,
"nbformat_minor": 5
}