界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
开发与工程 · SKILL.md

output_label_mapping

output_label_mapping

【开发与工程】Convert frozen source-model logits or predictions to target classes with AutoVP label mapping strategies.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: output_label_mapping description: Convert frozen source-model logits or predictions to target classes with AutoVP label mapping strategies.

Output Label Mapping

Use this skill for AutoVP FreqMap, IterMap-style refreshes, SemanticMap planning, or FullyMap linear logit mapping.

Inputs

Source predictions/logits, target labels, mapping multiplicity, optional class embeddings, and optional linear weights.

Outputs

Target-to-source mappings, target predictions, or mapped target logits.

Workflow

For FreqMap, count source predictions per target and choose deterministic top source labels. For IterMap, recompute FreqMap after prompt updates. For FullyMap, compute W L_s + b. Keep tie-breaking reproducible.

Validation

Run included deterministic tests.

我的收藏