mechanism_alignment_evaluation
Use this skill after a model-reprogramming recovery run to check whether the observed metric came from the intended mechanism: frozen source model, updated reprogramming parameters, reduced target loss, compatible target metadata, and optional alignment-distance decrease.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- mechanism_alignment_evaluation
- description
- Use this skill after a model-reprogramming recovery run to check whether the observed metric came from the intended mechanism: frozen source model, updated reprogramming parameters, reduced target loss, compatible target metadata, and optional alignment-distance decrease.
02 资源文件 2
03 SKILL.md 正文
mechanism_alignment_evaluation
When to use
Use this skill after a model-reprogramming recovery run to check whether the observed metric came from the intended mechanism: frozen source model, updated reprogramming parameters, reduced target loss, compatible target metadata, and optional alignment-distance decrease.
Do not use this skill to fine-tune or inspect an original paper repository. It is for reusable model-reprogramming mechanisms derived from the paper.
Inputs
- Small target-domain samples or source-output vectors.
- Declared source dimensions/classes and target dimensions/classes.
- Reprogramming parameters, masks, mappings, or training traces as appropriate.
Outputs
- Deterministic transformed vectors, target probabilities, training traces, or mechanism-check dictionaries.
- Explicit metadata sufficient for downstream validation of source-model immutability and trainable-parameter scope.
Workflow
- Validate dimensional assumptions from the model-reprogramming paper: target input dimension no larger than source input dimension and target label count no larger than source label count when using label mapping.
- Apply only the local transformation, mapping, training, or checking operation owned by this skill.
- Record numeric outputs and do not mutate frozen source model parameters.
- In recovery, save traces so the experiment gate can verify optimizer steps and mechanism checks.
Validation
Run python ../../../../Paper2Skills-Agent/src/packages/paper2skills-agent/src/paper2skills/skills/module-to-skill/scripts/validate_skill_tree.py <skill_dir> --run-tests from a suitable workspace, or directly run the tests in tests/.
Limitations
These utilities are deliberately small and deterministic. They validate the mechanism, not full-scale ImageNet, speech, biomedical, or language-model experiments.
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).