llm-html

CMDOP 技能 — 安装和使用通过 CMDOP agent:
Text
cmdop-skill install llm-html
Loading code...

LLM-优化的 HTML 清理：水合提取、令牌预算、多个输出格式。

安装

Bash

pip install llm-html

Loading code...

快速入门

Python

from llm_html import HTMLCleaner, CleanerConfig, OutputFormat

# 基本清理
cleaner = HTMLCleaner()
result = cleaner.clean(html)
print(f"减少：{result.stats.reduction_percent}%")

# 水合优先（从 Next.js、Nuxt 等提取 SSR 数据）
if result.hydration_data:
    data = result.hydration_data
else:
    cleaned = result.html

Loading code...

Convenience 函数

Python

from llm_html import clean, clean_to_json, clean_html, clean_for_llm

# 快速清理
result = clean(html)

# 获取 JSON 如果有 SSR 数据可用，否则清理的 HTML
data = clean_to_json(html)

# pipeline 有完全控制
result = clean_html(html, max_tokens=5000)
result = clean_for_llm(html, output_format="markdown")

Loading code...

输出格式

Python

from llm_html import to_markdown, to_aom_yaml, to_xtree

md = to_markdown(html)
aom = to_aom_yaml(html)
xtree = to_xtree(html)

Loading code...

Downsampling

令牌预算目标与 D2Snap 算法：

Python

from llm_html import downsample_html, estimate_tokens

tokens = estimate_tokens(html)
if tokens > 10000:
    html = downsample_html(html, target_tokens=8000)

Loading code...

Semantic Chunking

将大型页面分解为 LLM 尺寸的块：

Python

from llm_html import SemanticChunker, ChunkConfig

config = ChunkConfig(max_tokens=8000, max_items=20)
chunker = SemanticChunker(config)
result = chunker.chunk(soup)
for chunk in result.chunks:
    process(chunk.html)

Loading code...

Shadow DOM

将 Web 组件扁平化以使 LLM 可见：

Python

from llm_html import flatten_shadow_dom

flat = flatten_shadow_dom(html)

Loading code...

Helpers

Python

from llm_html import html_to_text, extract_links, extract_images, json_to_toon

text = html_to_text(html)
links = extract_links(html, base_url="https://example.com")
images = extract_images(html)
toon = json_to_toon({"key": "value"})

Loading code...

License

MIT

Text

cmdop-skill install llm-html

Bash

pip install llm-html

Python

from llm_html import HTMLCleaner, CleanerConfig, OutputFormat

# 基本清理
cleaner = HTMLCleaner()
result = cleaner.clean(html)
print(f"减少：{result.stats.reduction_percent}%")

# 水合优先（从 Next.js、Nuxt 等提取 SSR 数据）
if result.hydration_data:
    data = result.hydration_data
else:
    cleaned = result.html

Python

from llm_html import clean, clean_to_json, clean_html, clean_for_llm

# 快速清理
result = clean(html)

# 获取 JSON 如果有 SSR 数据可用，否则清理的 HTML
data = clean_to_json(html)

# pipeline 有完全控制
result = clean_html(html, max_tokens=5000)
result = clean_for_llm(html, output_format="markdown")

Python

from llm_html import to_markdown, to_aom_yaml, to_xtree

md = to_markdown(html)
aom = to_aom_yaml(html)
xtree = to_xtree(html)

Python

from llm_html import downsample_html, estimate_tokens

tokens = estimate_tokens(html)
if tokens > 10000:
    html = downsample_html(html, target_tokens=8000)

Python

from llm_html import SemanticChunker, ChunkConfig

config = ChunkConfig(max_tokens=8000, max_items=20)
chunker = SemanticChunker(config)
result = chunker.chunk(soup)
for chunk in result.chunks:
    process(chunk.html)

Python

from llm_html import flatten_shadow_dom

flat = flatten_shadow_dom(html)

Python

from llm_html import html_to_text, extract_links, extract_images, json_to_toon

text = html_to_text(html)
links = extract_links(html, base_url="https://example.com")
images = extract_images(html)
toon = json_to_toon({"key": "value"})

llm-html

llm-html

安装

快速入门

Convenience 函数

输出格式

Downsampling

Semantic Chunking

Shadow DOM

Helpers

License

评价