GPT-Image-2 Case · Multi-head Attention Heatmap

GPT-Image-2 reference case selected from 'Research Paper Figures' exhibits, suitable for exploring generated outcomes in 'multi-head attention heatmap'.

Author: AI Plus Lab

✦Results

◌Case Background

来自 GPT-Image2-Skill README 的精选展示条目，适合作为“研究论文图示 / 多头注意力热图”方向的站内参考案例。

⌘Prompt Content

Landscape 16:9 diagram, 4 attention heatmaps (2×2 grid), sharing 12 token inputs.

X-axis and Y-axis labeled markers (X-axis rotated 45°): "The", "quick", "brown", "fox", "jumped", "over", "the", "lazy", "dog", "near", "the", "river".

Four 12×12 cell panels, each titled:
"Layer 6, Head 3 — subject-verb" (highlighting cells between "fox" / "jumped")
"Layer 9, Head 7 — coreference" (highlighting cells between "the"(×2) / "river")
"Layer 11, Head 2 — prepositional" (highlighting cells between "over" / "dog", "near" / "river")
"Layer 14, Head 1 — sentence-final" (activity concentrated on the far right column)

Cell colors: dusty turquoise gradient, with deeper color indicating higher weight. Peak cells outlined with a 1px soft terracotta stroke. A shared vertical color bar on the far right, annotated with scales "0.0", "0.25", "0.5", "0.75", "1.0", label "attention weight".

Title: "Representative multi-head attention patterns in a 16-layer Transformer." Subtitle: "Examples from 4 heads, carefully chosen to showcase different head functionalities; inspired by Clark et al., 2019."

✎Outcome Notes

This case has been migrated from GPT-Image2-Skill README to AIPlusLab `/prompts`, for easy on-site retrieval, browsing, and reuse.