AGENT SKILL

linear_ood_calibration

Fit and inspect linear calibration between in-distribution and out-of-distribution accuracies.

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

01 前言 Front Matter

name
linear_ood_calibration
description
Fit and inspect linear calibration between in-distribution and out-of-distribution accuracies.

02 资源文件 2

03 SKILL.md 正文

Linear OOD Calibration

Use this skill after paired accuracy records have been validated. It implements the paper mechanism that OOD accuracy often lies on a line as ID accuracy varies across model variants.

Inputs

  • Validated records containing model_id, id_accuracy, and ood_accuracy.
  • Optional residual threshold for interpreting fit quality.

Outputs

  • Least-squares slope and intercept mapping ID accuracy to OOD accuracy.
  • Pearson correlation, mean absolute residual, and per-model predictions/residuals.
  • A deterministic JSON summary suitable for recovery logs.

Workflow

  1. Extract paired ID/OOD arrays in deterministic record order.
  2. Fit ordinary least squares with OOD accuracy as the dependent variable.
  3. Compute Pearson correlation and residuals.
  4. Return numeric evidence without deciding whether a soft-mode proxy is acceptable.
  5. Use residuals to inspect weak-shift or outlier behavior.

Validation

Run python scripts/fit_accuracy_line.py --input <validated_records.json> or run the included tests through validate_skill_tree.py --run-tests.

Limitations

This skill does not validate input records, choose the dataset, or claim full paper reproduction. It assumes records already obey the pair protocol.

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