sil_recovery_evaluation
Validate Self-Imitation Learning recovery evidence for source boundaries, executable metrics, and mechanism-faithful proxy checks.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- sil_recovery_evaluation
- description
- Validate Self-Imitation Learning recovery evidence for source boundaries, executable metrics, and mechanism-faithful proxy checks.
02 资源文件 2
- 脚本scripts/evaluate_sil_recovery.py3.0 KB
- 脚本tests/test_evaluate_recovery.py1.6 KB
03 SKILL.md 正文
SIL Recovery Evaluation
Use this skill after running a SIL recovery experiment. It is designed for soft-mode reduced recoveries and full actor-critic recoveries that need auditable acceptance checks before analysis.
Do not use this skill to waive missing executable evidence; it should reject missing command logs, missing optimizer-step evidence, or original-repository source leakage.
Inputs
- Recovery result JSON.
- Experiment command log JSON.
- Generated skill invocation log JSON.
- Source manifest JSON.
- Optional training trace JSON.
Outputs
- Validation JSON with
ok,errors,warnings, and metric summary. - Mechanism checklist suitable for Distiller analysis.
Workflow
- Check that source manifest entries do not include the original source repository during recovery.
- Check that a successful command produced the recovery result.
- Check that generated skills were invoked or cross-checked.
- Check that proxy mechanism booleans include replay insertion, positive-advantage loss, optimizer step, and parameter change.
- Report metric and loss-change evidence.
Validation
Run:
python tests/test_evaluate_recovery.py
The test validates a minimal passing proxy result and a failing result with missing optimizer evidence.
Limitations
- This is a SIL-specific supplement to Distiller's generic
validate_recovery_experiment.py; run both when finishing a recovery.
Source-Boundary Negative Test
The recovery source manifest should not include the original implementation repository path, even as a convenience reference. Record source-boundary status without embedding forbidden paths in recovery evidence.
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).