python3 -m venv ~/venvs/docling
source ~/venvs/docling/bin/activate   # Windows: ~/venvs/docling/Scripts/activate
pip install docling
docling --version

python -m venv $env:USERPROFILE\venvs\docling
.\$env:USERPROFILE\venvs\docling\Scripts\Activate.ps1
pip install docling

# 安装 uv 后，用 uvx 直接试用，无需建环境
uvx --from docling docling https://arxiv.org/pdf/2206.01062

# 或在项目中作为依赖安装
uv add docling
uv run docling --version

docling https://arxiv.org/pdf/2206.01062

docling-tools models download ./docling-models
# 也可下载指定的 HuggingFace 仓库：
docling-tools models download-hf-repo ./docling-models ibm-granite/granite-docling-258M

export HF_ENDPOINT=https://hf-mirror.com
docling-tools models download ./docling-models

docling --artifacts-path ./docling-models --device cpu report.pdf

docker run -d --name docling-serve \
  -p 5001:5001 \
  -e DOCLING_SERVE_ENABLE_UI=1 \
  quay.io/docling-project/docling-serve

pip install docling-serve
docling-serve run --enable-ui

git clone https://github.com/docling-project/docling.git
cd docling
uv sync            # 或 pip install -e .
uv run docling --version

# 离线 + CPU + 中文 OCR + 快速表格，输出 Markdown 和 JSON
docling convert report.pdf \
  --artifacts-path ./docling-models \      # 离线模型目录
  --device cpu \                            # 强制 CPU
  --ocr-engine rapidocr \                   # 选 OCR 引擎
  --ocr-lang iso:zh-Hans \                  # 简体中文
  --table-mode fast \                       # 表格用快速模式
  --to md --to json \                       # 双格式输出
  --output ./out                            # 输出目录

docling-serve run --enable-ui

# /etc/systemd/system/docling-serve.service
[Unit]
Description=Docling Serve API
After=network.target

[Service]
User=docling
WorkingDirectory=/opt/docling
Environment="DOCLING_SERVE_ENABLE_UI=1"
Environment="DOCLING_ARTIFACTS_PATH=/opt/docling/models"
ExecStart=/opt/docling/venv/bin/docling-serve run
Restart=always

[Install]
WantedBy=multi-user.target

sudo systemctl daemon-reload
sudo systemctl enable --now docling-serve

services:
  docling-serve:
    image: quay.io/docling-project/docling-serve:latest
    ports:
      - "5001:5001"
    environment:
      - DOCLING_SERVE_ENABLE_UI=1
    restart: unless-stopped

# 用 NSSM 把 docling-serve 注册为 Windows 服务
nssm install DoclingServe C:\docling\venv\Scripts\docling-serve.exe run
nssm set DoclingServe AppEnvironmentExtra DOCLING_SERVE_ENABLE_UI=1
nssm start DoclingServe

docling-serve run --help   # 查看端口相关参数（以官方帮助为准，默认 5001）

location /docling/ {
    proxy_pass http://127.0.0.1:5001/;
    proxy_set_header Host $host;
    client_max_body_size 200m;
}

docling convert paper.pdf --to md --to json --output ./out
# 输出：./out/paper.md 与 ./out/paper.json

docling convert scan.pdf --ocr --ocr-mode full_page --to md

docling convert --pipeline vlm --vlm-model granite_docling https://arxiv.org/pdf/2206.01062

from docling.document_converter import DocumentConverter

source = "https://arxiv.org/pdf/2408.09869"  # 本地路径或 URL
converter = DocumentConverter()
result = converter.convert(source)
print(result.document.export_to_markdown())
print(result.document.export_to_dict())   # DoclingDocument 无损 JSON

from docling.document_converter import DocumentConverter

conv = DocumentConverter()
doc = conv.convert("report.pdf").document
for table in doc.tables:
    df = table.export_to_dataframe()   # 导出为 pandas DataFrame
    print(df.head())

docling convert handbook.pdf --to chunks --chunks-type hybrid --output ./chunks

curl -X 'POST' 'http://localhost:5001/v1/convert/source' \
  -H 'Content-Type: application/json' \
  -d '{"sources": [{"kind": "http", "url": "https://arxiv.org/pdf/2501.17887"}]}'

curl -X POST "http://localhost:5001/v1/convert/source/async" \
  -H "Content-Type: application/json" \
  -d '{"http_sources": [{"url": "https://arxiv.org/pdf/2501.17887"}]}'

pip install docling-llamaindex       # LlamaIndex 官方集成
pip install langchain-docling        # LangChain 官方集成

# LlamaIndex 示例（RAG）
from docling_llamaindex.docling_reader import DoclingReader
from llama_index.core import SimpleDirectoryReader

reader = DoclingReader()
loader = SimpleDirectoryReader(
    input_dir="./docs", file_extractor={".pdf": reader}
)
documents = loader.load_data()

# LangChain 示例
from langchain_docling import DoclingLoader

loader = DoclingLoader(file_path="./docs/manual.pdf", export_type="md")
docs = loader.load()

docling --version
pip install -U docling docling-serve   # 升级到最新
pip install docling==2.142.1           # 锁定指定版本

docker pull quay.io/docling-project/docling-serve:latest
docker stop docling-serve && docker rm docling-serve
docker run -d --name docling-serve -p 5001:5001 \
  -e DOCLING_SERVE_ENABLE_UI=1 quay.io/docling-project/docling-serve

# 回滚：改用旧版本 tag
docker run -d --name docling-serve -p 5001:5001 \
  quay.io/docling-project/docling-serve:<旧版本tag>

pip uninstall docling docling-serve
rm -rf ~/.cache/docling ~/.cache/huggingface   # 模型缓存
rm -rf ~/venvs/docling                          # 虚拟环境
docker rm -f docling-serve                      # 容器
docker rmi quay.io/docling-project/docling-serve # 镜像

# 安装
pip install docling                       # 基础安装
uvx --from docling docling <文件>          # 免安装试用
pip install docling-serve                 # API 服务

# 模型
docling-tools models download ./models    # 预下载 artifacts
export HF_ENDPOINT=https://hf-mirror.com  # HF 镜像

# 常用转换
docling convert a.pdf --to md             # PDF → Markdown
docling convert a.pdf --to md --to json --output out/
docling convert scan.pdf --ocr --ocr-mode full_page
docling convert --pipeline vlm --vlm-model granite_docling a.pdf
docling convert a.pdf --to chunks         # RAG 分块
docling convert --page-range 1-4 --table-mode accurate a.pdf

# 服务
docling-serve run --enable-ui             # 前台启动（端口 5001）
curl -X POST localhost:5001/v1/convert/source \
  -H 'Content-Type: application/json' \
  -d '{"sources":[{"kind":"http","url":"https://example.com/a.pdf"}]}'

# 排错
docling convert a.pdf -v                  # info 日志
docling convert a.pdf -vv                 # debug 日志
docling convert a.pdf --device cpu --num-threads 2