AGENT SKILL

self-training

Operate Transfer Learning Library semi-supervised and self-training workflows using TLLib pseudo-label, consistency, teacher, thresholding, and debiasing APIs.

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

01 前言 Front Matter

name
self-training
license
MIT
metadata
{ "disco-role": "operating" }
description
Operate Transfer Learning Library semi-supervised and self-training workflows using TLLib pseudo-label, consistency, teacher, thresholding, and debiasing APIs.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

Self-Training Skill

Use this sub-skill when a user needs TLLib guidance for semi-supervised learning or self-training on image classification tasks: pseudo-labeling, Pi Model, Mean Teacher, UDA, FixMatch, FlexMatch, Self-Tuning, DST, DebiasMatch-style thresholding, Noisy Student, or MCC-like self-training regularizers.

Route here for

  • Confidence-based pseudo-label selection, consistency losses, weak/strong augmentation training loops, EMA teacher/student updates, dynamic thresholding, class-balance regularization, DST worst-case estimation, and Self-Tuning queues.
  • Choosing between Pseudo Label, Pi Model, Mean Teacher, UDA, FixMatch, FlexMatch, Noisy Student, Self-Tuning, DebiasMatch-style debiasing, DST, and MCC-style regularization.
  • Safe CPU checks for tllib.self_training losses/helpers without datasets, checkpoints, GPUs, or benchmark training.
  • MoCo-style checkpoint conversion planning before SSL fine-tuning.

Route elsewhere

  • Datasets, image-list formats, class-balanced labeled subsets, transforms, model factories, metrics, and utility loaders: vision-data-models.
  • Task adaptation / fine-tuning regularizers or checkpoint use after conversion: task-generalization.
  • Domain-adaptation tasks that use MCC on target-domain predictions: domain-adaptation.

Operating sequence

  1. Confirm the task is semi-supervised/self-training: a small labeled split plus a larger unlabeled split, or a teacher/student self-training loop. If the user is adapting across source/target domains, route DA-specific pieces to domain-adaptation.
  2. Confirm data/model readiness through vision-data-models. Full SSL training needs datasets, class-balanced labeled sampling, weak/strong augmentations, pretrained backbones or checkpoints, and usually CUDA; this sub-skill only claims CPU component-level verification.
  3. Choose the method family:
    • Simple pseudo labels: ConfidenceBasedSelfTrainingLoss.
    • Prediction consistency: L2ConsistencyLoss / ConsistencyLoss.
    • Teacher/student averaging: EMATeacher plus a consistency loss.
    • Weak/strong augmentation: UDA or FixMatch-style losses.
    • Curriculum thresholding: DynamicThresholdingModule for FlexMatch.
    • Self-supervised or contrastive adaptation: Self-Tuning and MoCo checkpoint conversion.
    • Debiasing: DST heads/losses or DebiasMatch-style class-balance/threshold checks.
  4. Use the bundled references for API signatures, workflow recipes, checkpoint conversion, and troubleshooting. Do not copy or execute repository benchmark trainers as runtime helpers.
  5. For installation/API sanity, run scripts/tllib_self_training_smoke.py from any directory where tllib is installed.

References

Verification status

The bundled smoke script exercises tiny CPU tensors for pseudo-label selection, consistency, EMA teacher updates, UDA weak/strong loss, MCC, FlexMatch dynamic thresholds, class-balance loss, and DST worst-case estimation. It does not run benchmark training, download datasets, download pretrained checkpoints, perform MoCo conversion on external files, or verify optional timm/CUDA training stacks.

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