AGENT SKILL

gedi_posterior

Compute GeDi generative-discriminator class posteriors from control and anti-control log probabilities using stable Bayes-rule normalization.

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

01 前言 Front Matter

name
gedi_posterior
description
Compute GeDi generative-discriminator class posteriors from control and anti-control log probabilities using stable Bayes-rule normalization.

02 资源文件 2

03 SKILL.md 正文

GeDi Posterior

When to use

Use this skill when implementing or checking GeDi-style controlled generation or training and you have class-conditional sequence log probabilities for desired and undesired control codes. Do not use it for ordinary discriminative classifiers that do not compare generative likelihoods.

Inputs

  • Per-class sequence or token log probabilities.
  • Sequence length for length normalization.
  • Optional class biases/log priors.
  • alpha normalization scale.

Outputs

  • Stable posterior probabilities over classes.
  • Desired-class probability and diagnostic logits.

Workflow

  1. Sum token log probabilities per class if token-level values are provided.
  2. Normalize each class log-likelihood by alpha / length.
  3. Add class bias terms.
  4. Apply stable log-softmax; for binary cases, sigmoid of the logit difference is equivalent.
  5. Use the returned posterior as input to decoding or discriminative training.

Validation

Run:

python tests/test_posterior.py

Limitations

This skill computes the GeDi posterior only. It does not run a language model, tokenize text, or select next tokens.

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