AGENT SKILL

normalization_configuration

Configure normalization layers for Tent by exposing only affine modulation parameters and target batch statistics.

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

01 前言 Front Matter

name
normalization_configuration
description
Configure normalization layers for Tent by exposing only affine modulation parameters and target batch statistics.

02 资源文件 2

03 SKILL.md 正文

Normalization Configuration for Tent

Use this skill before running entropy-minimizing test-time adaptation on a model with normalization layers. It is appropriate for creating a Tent configuration report, checking a proxy model, or comparing against normalization-only adaptation. Do not use it to authorize updating all model weights unless the run is an explicit ablation outside standard Tent.

Inputs

  • A model inventory JSON with modules, parameter names, layer types, and affine/statistics fields.
  • Optional flags for target batch statistics and whether running statistics should be disabled.

Outputs

  • Trainable parameter names limited to normalization affine scale and shift.
  • A configured module inventory with target-batch statistics enabled for normalization layers.
  • Validation errors for missing normalization layers or accidental full-model training.

Workflow

  1. Set the model-level mode to training for adaptation.
  2. Disable gradients for all parameters by default.
  3. For each batch normalization descriptor, enable gradients on weight and bias only.
  4. Set track_running_stats false when using target-batch-only statistics.
  5. Return a report with trainable parameters and configuration checks.
  6. Reject configurations where no normalization affine parameters are available.

Validation

Run python scripts/configure_norm.py --self-test. The test constructs a tiny inventory and confirms that only batch-normalization affine parameters are trainable while classifier parameters remain frozen.

Limitations

The script works on portable inventory dictionaries. For live frameworks such as PyTorch, use this report as a contract and apply the same logic to actual modules.

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