rnd_observation_normalization
Normalize observations for Random Network Distillation with running statistics and clipping before target and predictor feature computation.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- rnd_observation_normalization
- description
- Normalize observations for Random Network Distillation with running statistics and clipping before target and predictor feature computation.
02 资源文件 2
- 脚本scripts/normalization.py3.4 KB
- 脚本tests/test_normalization.py401 B
03 SKILL.md 正文
RND Observation Normalization
Use this skill when implementing or validating Random Network Distillation (RND) pipelines that feed observations into a frozen random target and a trainable predictor. Do not use it as a generic image augmentation skill; its purpose is to preserve the paper's scale-control mechanism.
Inputs
- A batch of numeric observations represented as lists of equal-length vectors.
- Optional running statistics with
mean,var, andcountfields. - Optional clipping bounds, defaulting to the paper's
[-5, 5]range.
Outputs
- Normalized and clipped observation vectors.
- Updated running statistics that must be shared by target and predictor paths.
Workflow
- Merge incoming batch moments with existing running moments using a numerically stable parallel variance update.
- Normalize each dimension with
(x - mean) / sqrt(var + eps). - Clip normalized values after whitening.
- Feed exactly the same normalized batch to both RND target and predictor modules.
Validation
Run python scripts/normalization.py --self-test or validate the skill tree with tests enabled. The self-test checks running-stat updates, finite normalization, and clipping bounds.
Limitations
The script handles vector observations for deterministic recovery and tests. Image tensors should be flattened or batched into vectors before use, while preserving the same running-stat semantics.
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).