soft_regularization
Compute soft movement-pruning sigmoid sparsity penalties and gradients for threshold-mask recovery checks.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- soft_regularization
- description
- Compute soft movement-pruning sigmoid sparsity penalties and gradients for threshold-mask recovery checks.
02 资源文件 3
03 SKILL.md 正文
Soft Movement Sparsity Regularization
Use this skill when implementing or auditing soft movement pruning. It computes the paper penalty lambda_mvp * sum(sigmoid(S)) and the corresponding score-gradient contribution.
Inputs
scores: flat or nested numeric scores.lambda_mvp: nonnegative regularization strength.- Optional
thresholdandlrfor before/after keep-ratio diagnostics.
Outputs
Penalty value, gradient contribution with the same shape as scores, sigmoid mass, and optional threshold keep-ratio diagnostics.
Workflow
- Compute stable sigmoid values.
- Sum sigmoid scores and multiply by
lambda_mvp. - Compute
lambda_mvp * sigmoid(S) * (1 - sigmoid(S)). - Optionally simulate a regularization-only gradient step to show scores are pushed downward.
Validation
Run python tests/test_soft_regularization.py. Tests check zero-lambda behavior, sigmoid-at-zero values, and non-increasing scores under the penalty update.
Limitations
This skill supplies only the regularizer component; task loss and score masks must be handled by other skills.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).