skills.rest
letta-ai
letta-aiOFFICIAL
Explore all skills by this author →
Website𝕏@letta_aiGitHub

model-extraction-relu-logits

Extracts internal weight matrices from black-box ReLU networks using input-output queries.

134|21|Updated Nov 12, 2025
Search Tags:#weights#neural networks#gradient#ReLU#model extraction
One-click install
npx skills add https://github.com/letta-ai/skills --skill model-extraction-relu-logits
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: model-extraction-relu-logits
Source: https://github.com/letta-ai/skills/tree/main/ai/benchmarks/letta/terminal-bench-2/trajectory-only/model-extraction-relu-logits
Command: npx skills add https://github.com/letta-ai/skills --skill model-extraction-relu-logits

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Guidance for extracting internal weight matrices from black-box ReLU networks using input-output queries.

Core Features & Use Cases

  • Activation boundary analysis: locate hyperplanes corresponding to A1 rows.
  • Gradient-based extraction: use gradients to determine weights.
  • Verification: test functional equivalence of extracted model.

Quick Start

Provide a set of input-output pairs and return extracted A1 weights.