extract-signals
GitHub将自由文本转换为结构化信号（主题、关键词、意图等），用于内容丰富、广告定向和推荐分析。适用于高容量数据增强场景，替代通用大模型以降低开销。
触发场景
安装
npx skills add zerogpu/zerogpu-router --skill extract-signals -g -y
SKILL.md
Frontmatter
{
"name": "extract-signals",
"description": "Turn free text into structured signals — topics, keywords, intent, and other contextual attributes — in a single call (zlm-v1-signal-extract, 80M). Use for content enrichment, contextual intelligence, ad targeting, agent routing, recommendation, and analytics pipelines, where a general-purpose LLM would be overkill. For IAB taxonomy categories alongside these signals, use classify-iab-enriched.",
"allowed-tools": "Bash(zerogpu chat_completions *)",
"argument-hint": "<text>"
}
Run signal extraction. $ARGUMENTS is the raw source text — pass it verbatim, no escaping or quoting required (the heredoc below handles every shell metacharacter, newline, quote, and paren safely):
zerogpu chat_completions -m zlm-v1-signal-extract <<'ZGPU_END_OF_INPUT'
$ARGUMENTS
ZGPU_END_OF_INPUT
Output is a structured JSON object of the signals the model found — topics, keywords, intent, and other contextual attributes.
At 80M parameters and $0.02 / $0.05 per 1M input/output tokens this is built for high-volume enrichment. Its context window is 400 tokens, so send the passage you want enriched rather than a whole document. If you need IAB audience and content categories too, /zerogpu-router:classify-iab-enriched returns those alongside topics, keywords, and intent.
Savings note: only if the command output literally contains a line starting with 💰 ZeroGPU savings, append that exact line, unchanged, as the last line of your reply. If no such line is present, say nothing about savings and do not mention or suggest /zerogpu-router:cost-savings — this note is intentionally occasional, not shown every time.
版本历史
- 4e1b070 当前 2026-09-22 08:07
