AGENT SKILL

conditional_nce_training

Train a tiny residual sequence-energy model with the paper's conditional NCE objective on positive and fixed-LM negative continuations.

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

01 前言 Front Matter

name
conditional_nce_training
description
Train a tiny residual sequence-energy model with the paper's conditional NCE objective on positive and fixed-LM negative continuations.

02 资源文件 2

03 SKILL.md 正文

Conditional NCE Training

Use this skill when a recovery or implementation needs to exercise the paper's training mechanism: positive human continuations should receive low energy and fixed language-model continuations should receive high energy for the same prefix.

Do not use this skill to update the base language model or to mine negatives from the residual model. The noise distribution is the fixed LM proposal.

Inputs

  • A shared prefix.
  • Positive continuation records.
  • Negative continuation records sampled or designated as fixed-LM proposal outputs.
  • Feature weights for a scalar residual energy function.
  • Learning rate and update count.

Outputs

  • NCE loss before and after training.
  • Parameters before and after training.
  • Positive and negative energy diagnostics.
  • A validator-compatible trace showing optimizer state or parameter changes.

Workflow

  1. Encode continuations into deterministic lexical or model-derived features. Include both positive evidence features and artifact features; repeated "quality" terms must not be enough to bypass repetition or coherence penalties.
  2. Compute scalar energy as a linear function of those features plus bias.
  3. Minimize softplus(E_positive) + softplus(-E_negative) averaged over examples.
  4. Update only energy parameters.
  5. Record the energy gap and loss change; reduced recovery should mark this as reduced training, not full pretrained model training.

Validation

Run:

python scripts/nce_train.py --demo
python -m pytest tests

The tests verify loss decrease, parameter changes, positive energy lower than negative energy after a bounded deterministic run, and robustness against an adversarial negative that repeats otherwise positive lexical terms.

Limitations

This skill can run a small deterministic optimizer for recovery evidence. It does not reproduce the paper's large transformer energy network unless a caller supplies that runtime separately and records the environment evidence.

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