yzj623/pplab/activation-calibration
Claude Code and agent skill by yzj623/pplab
- community
Checking that the SKILL.md file is still in the repository.
Estimate per-channel activation norms for Wanda outlier-aware pruning from calibration tensors.
Install
npx vskill@latest install yzj623/pplab/activation-calibrationSKILL.md
The same SKILL.md was published earlier as vectorspacelab/arex-skill/activation-calibration.
Activation Calibration
Use this skill when implementing Wanda-style pruning and you need input-channel activation scales before computing weight importance. Do not use it for methods that compare weights only by magnitude.
Inputs
- A JSON file containing
batches, a list of 2D numeric arrays shaped samples/tokens by input channels, or one 2D array. - Optional norm type:
l2is the Wanda default.
Outputs
activation_norms: one non-negative scale per input channel.- Metadata with channel count, row count, norm type, and outlier-channel ranking.
Workflow
- Flatten calibration examples across batch/token rows while preserving the final feature dimension.
- Compute the selected norm independently for every input channel; Wanda uses L2 because it smoothly captures high-magnitude outlier features.
- Keep zero-norm channels as zero, not NaN or epsilon-inflated values.
- Save norms and metadata for the row-pruning skill.
Validation
Run python tests/test_activation_calibration.py or validate the skill tree with --run-tests.
Limitations
This skill estimates scales only; it does not choose prune masks or evaluate perplexity.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingResearchAgent-Skill/skill_library/paperbench/adaptive-pruning/skill/wanda_outlier_pruning/activation_calibration/SKILL.md
Popularity
- 7-day trend score
- 5
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/yzj623/pplab/activation-calibration)