jsrl_switching_rollout_curriculum
Build Jump-Start Reinforcement Learning rollouts that switch from guide-policy control to exploration-policy control under curriculum or random guide-step schedules.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- jsrl_switching_rollout_curriculum
- description
- Build Jump-Start Reinforcement Learning rollouts that switch from guide-policy control to exploration-policy control under curriculum or random guide-step schedules.
02 资源文件 3
03 SKILL.md 正文
JSRL Switching Rollout Curriculum
When To Use
Use this skill to collect trajectories for JSRL or to test whether a recovery harness correctly implements the paper's two-policy handoff. It is useful for full RL environments and for reduced sparse-reward proxy environments.
Inputs
- Environment with
reset()andstep(action). - Guide-policy and exploration-policy callables.
- Horizon
H, selected guide-steph, and optional guide-step schedule. - Threshold
βused to advance curriculum stages.
Outputs
- Trajectory records with state, action, reward, next state, done flag, and
controllerequal toguideorexploration. - Handoff summary with selected
h, guide action count, exploration action count, total reward, and success flag. - Curriculum cursor updates for threshold-based advancement.
Workflow
- Clamp the selected guide-step to the valid horizon.
- Roll out the guide-policy while
t < h. - Roll out the exploration-policy after the handoff until termination or horizon.
- Append controller labels to every transition.
- Advance the curriculum only after evaluation meets or exceeds
β. - For ablations, sample guide steps from the same schedule without changing the data schema.
Validation
Run:
python scripts/switching_rollout.py --demo
python -m pytest tests
The demo prints a small trajectory and handoff summary.
Limitations
This skill does not implement a complete RL algorithm. It only owns the rollout and curriculum contract; policy/value updates belong in the value exploration update skill.
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).