AGENT SKILL

rnd_intrinsic_reward

Compute and test Random Network Distillation intrinsic rewards from predictor error against a fixed deterministic target feature map.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
rnd_intrinsic_reward
description
Compute and test Random Network Distillation intrinsic rewards from predictor error against a fixed deterministic target feature map.

02 资源文件 2

03 SKILL.md 正文

RND Intrinsic Reward

Use this skill when implementing or auditing an RND novelty bonus. Do not use it for forward-dynamics curiosity or stochastic target prediction.

Inputs

  • Observation vectors as JSON arrays or Python lists of floats.
  • Fixed target parameters and trainable predictor parameters.
  • Optional normalized observations from a separate normalizer.

Outputs

  • Per-observation mean squared prediction errors used as intrinsic rewards.
  • Distillation loss over a batch.
  • Updated predictor parameters when training is requested.

Workflow

  1. Keep target parameters immutable for the whole experiment.
  2. Compute target and predictor features from the same observation batch.
  3. Use mean squared error across feature dimensions as intrinsic reward.
  4. Train only predictor parameters on visited observations.
  5. Validate novelty by comparing trained/visited error with held-out rare-state error.

Validation

Run python tests/test_rnd_intrinsic_reward.py from this skill directory.

Limitations

The included script is a deterministic linear RND proxy for bounded recovery and tests; large Atari recovery still needs a full neural policy/runtime.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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).

★ 351