deberta_recovery_experiment_harness
Orchestrate a source boundary clean DeBERTa reduced recovery harness that records executable evidence and mechanism checks.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- deberta_recovery_experiment_harness
- description
- Orchestrate a source boundary clean DeBERTa reduced recovery harness that records executable evidence and mechanism checks.
02 资源文件 2
- 脚本scripts/recovery_harness.py2.1 KB
- 脚本tests/test_recovery_harness.py1.2 KB
03 SKILL.md 正文
DeBERTa Recovery Experiment Harness
Use this skill after the DeBERTa module skills and environment handoff exist. It helps a recovery script copy the module-plan target, record allowed sources, log generated skill invocations, and decide whether a soft-mode reduced proxy is valid. Do not use it to justify a proxy when hard recovery mode is active.
Inputs
- Attempt directory.
- Generated skills root.
module_plan.json.environment/runtime_handoff.json.- Recovery mode from
run_manifest.json.
Outputs
- Source manifest structure.
- Target consistency checks.
- Invocation-log entries for generated skills.
- Validator-friendly mechanism-check fields.
Workflow
- Read the module-plan target and keep it as authoritative.
- Read the runtime handoff and require blockers before selecting reduced recovery.
- Build a source manifest that excludes the original repository.
- Record one invocation entry for each generated skill that is imported, called, cross-checked, or marked not applicable.
- Mark full model booleans false when the model stack is unavailable.
Validation
The bundled tests verify target copying, source manifest construction, and reduced-recovery eligibility under soft mode with runtime blockers.
Limitations
This skill is a harness helper. The actual experiment command must still run a recovery script that produces metrics, training traces, and Distiller recovery-gate 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).