reduced_cic_recovery_harness
Run a bounded soft-mode CIC proxy experiment that invokes generated batching, contrastive loss, and entropy reward skills.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- reduced_cic_recovery_harness
- description
- Run a bounded soft-mode CIC proxy experiment that invokes generated batching, contrastive loss, and entropy reward skills.
02 资源文件 2
- 脚本scripts/run_reduced_cic_recovery.py7.9 KB
- 脚本tests/test_reduced_harness.py1.3 KB
03 SKILL.md 正文
Reduced CIC Recovery Harness
Use this skill when full URLB CIC pretraining is blocked or too expensive and soft-mode recovery allows a declared mechanism-faithful proxy. The harness must invoke the generated CIC module scripts and produce validator-compatible recovery artifacts.
Inputs
attempt_dir: initialized Paper2Skills attempt directory.skill_root: directory containing the generated CIC module skills.runtime_handoff: environment handoff JSON from the prepare-recovery-environment stage.- Optional seed and batch dimensions for deterministic reduced recovery.
Outputs
recovery/recovery_result.json.recovery/logs/generated_data_item.json.recovery/logs/training_trace.json.recovery/logs/generated_skill_invocations.json.- Supporting metrics and mechanism checks.
Workflow
- Load the module-plan recovery target and runtime handoff.
- Import the generated transition batching, CIC loss, and particle entropy scripts from
skill_root. - Build a deterministic transition-skill proxy batch.
- Compute CIC contrastive loss and kNN entropy rewards.
- Run one dependency-free optimizer step on tiny linear encoders.
- Save recovery artifacts with explicit proxy and reduced-training declarations.
- Record invocation evidence for each core generated skill.
Validation
Run:
python scripts/run_reduced_cic_recovery.py --attempt-dir <attempt_dir> --skill-root <skill_root>
python tests/test_reduced_harness.py
Limitations
This harness does not claim full URLB results. It is a soft-mode proxy that validates the CIC mechanism under bounded runtime constraints.
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).