Name: structured_sparsity_patterns
Author: yzj623

Imported from yzj623/pplab (basicagent-skill/skill_library/paperbench/adaptive-pruning/skill/wanda_outlier_pruning/structured_sparsity_patterns/SKILL.md). Install upstream with npx skills add yzj623/pplab --skill structured_sparsity_patterns. Copyright stays with the author.

Structured Sparsity Patterns

Use this skill when a Wanda recovery or implementation needs 2:4, 4:8, or other N:M pruning masks. It operates on score matrices after activation-aware scoring has already been computed.

Inputs

Score matrix shaped outputs by inputs.
prune_n: number of entries to prune in each group.
prune_m: group width along the input-channel dimension.

Outputs

Boolean mask with exactly prune_n true values per complete group in each row.
Metadata describing complete groups, ignored tail width, and achieved sparsity.

Workflow

Validate 0 < prune_n <= prune_m.
Split each row into consecutive groups of width prune_m.
In each complete group, mark the prune_n lowest scores for pruning with deterministic tie-breaking by column index.
Leave incomplete tail groups unpruned unless the caller pads them explicitly.

Validation

Tests assert exact group counts and unpruned tail behavior.

Limitations

The skill only creates masks. Applying them and checking model quality belongs to an evaluation skill.

# after Install: the listing is in your workspace's routing pool - a plan picks it for its slot curl -s -X POST https://api.opensmartroute.ai/api/v1/route -H 'Authorization: Bearer $OSR_API_KEY' -H 'Content-Type: application/json' -d '{"text": "...", "plan": true}'

{ "ocm": "1", "id": "yzj623-pplab-structured-sparsity-patterns", "kind": "skill", "name": "structured_sparsity_patterns", "description": "Build deterministic N:M semi-structured pruning masks from Wanda or other importance scores.", "publisher": "yzj623", "version": "1.0.0", "capabilities": { "domains": [ "general" ], "tags": [ "skill-md", "github" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Build deterministic N:M semi-structured pruning masks from Wanda or other importance scores." ], "primary": false, "metadata": { "source": { "provider": "github", "repository": "https://github.com/yzj623/pplab", "path": "basicagent-skill/skill_library/paperbench/adaptive-pruning/skill/wanda_outlier_pruning/structured_sparsity_patterns/SKILL.md", "ref": "da871e009e7729366f67b44a9c143f193b59fb76", "url": "https://github.com/yzj623/pplab/blob/da871e009e7729366f67b44a9c143f193b59fb76/basicagent-skill/skill_library/paperbench/adaptive-pruning/skill/wanda_outlier_pruning/structured_sparsity_patterns/SKILL.md", "key": "yzj623/pplab/basicagent-skill/skill_library/paperbench/adaptive-pruning/skill/wanda_outlier_pruning/structured_sparsity_patterns/SKILL.md" } }, "instructions": "# Structured Sparsity Patterns\n\nUse this skill when a Wanda recovery or implementation needs 2:4, 4:8, or other N:M pruning masks. It operates on score matrices after activation-aware scoring has already been computed.\n\n## Inputs\n- Score matrix shaped outputs by inputs.\n- `prune_n`: number of entries to prune in each group.\n- `prune_m`: group width along the input-channel dimension.\n\n## Outputs\n- Boolean mask with exactly `prune_n` true values per complete group in each row.\n- Metadata describing complete groups, ignored tail width, and achieved sparsity.\n\n## Workflow\n1. Validate `0 < prune_n ", "cost": { "context_tokens": 267 } }

structured_sparsity_patterns

About