proxy_recovery_evaluation
Decide whether a declared soft-mode proxy exercises the Accuracy on the Line mechanism.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- proxy_recovery_evaluation
- description
- Decide whether a declared soft-mode proxy exercises the Accuracy on the Line mechanism.
02 资源文件 2
- 脚本scripts/evaluate_proxy.py1.5 KB
- 脚本tests/test_proxy_recovery_evaluation.py1009 B
03 SKILL.md 正文
Proxy Recovery Evaluation
Use this skill after a recovery harness has produced calibration metrics. It evaluates whether a reduced or proxy experiment is acceptable evidence under soft mode. Do not use it to fabricate metrics; it consumes executable experiment outputs.
Inputs
- A calibration result with
pearson_r, slope/intercept, and residuals. - A declared target threshold, usually from
module_plan.json.fast_recovery_target.paper_value. - Mechanism checks covering paired records, line fitting, residual inspection, source boundary, and proxy declaration.
Outputs
accepted_proxy,metric_gap,mechanism_ok, and human-readable reasons.- A compact JSON decision that analysis can inspect.
Workflow
- Confirm the recovery is explicitly declared as proxy evidence.
- Check that all required mechanism booleans are true.
- Compare the observed Pearson correlation against the declared threshold.
- Report metric gap and reasons without upgrading proxy evidence to full reproduction.
Validation
Run the included tests or call python scripts/evaluate_proxy.py --calibration <file> --checks <file> --threshold 0.95.
Limitations
This skill accepts soft-mode proxy evidence only. Hard-mode runs or full benchmark claims require real benchmark results rather than this reduced decision.
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).