jsrl_guide_policy_contract
Validate a Jump-Start Reinforcement Learning guide-policy interface and better-than-random progress assumptions before using it for roll-in.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- jsrl_guide_policy_contract
- description
- Validate a Jump-Start Reinforcement Learning guide-policy interface and better-than-random progress assumptions before using it for roll-in.
02 资源文件 3
03 SKILL.md 正文
JSRL Guide Policy Contract
When To Use
Use this skill when a recovery or implementation needs to decide whether a prior policy is suitable as the fixed JSRL guide-policy. It is appropriate for scripted, imitation-trained, offline-RL, or manually engineered guides. Do not use it to train the guide; it only wraps and validates one.
Inputs
- A callable or table-like guide-policy that maps observations/states to actions.
- An environment or small deterministic probe exposing legal actions and progress-to-goal information.
- Optional baseline/random policy for comparison.
Outputs
- Validated action selections from
act. - A guide quality report with success count, average progress, and
useful: true/false. - Warnings for stationary, invalid-action, or adversarial guides.
Workflow
- Wrap the guide behind a minimal
act(state)function. - Check every returned action against the environment's legal actions.
- Run bounded rollouts and compare progress or success against a baseline.
- Treat a guide as useful only if it improves progress or success enough to create good handoff states.
- Record limitations when the guide is related-task, stochastic, or suboptimal.
Validation
Run:
python scripts/guide_policy_contract.py --demo
python -m pytest tests
The script is standard-library only and can also be imported by recovery harnesses.
Limitations
This skill does not prove the guide is optimal. It only checks the paper's minimum JSRL assumption: the guide can act from observations and is better than a random or stationary baseline on a bounded probe.
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).