AGENT SKILL

training-workflows

Lightly SSL training recipes for PyTorch, Lightning, and gated distributed variants.

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

01 前言 Front Matter

name
training-workflows
license
MIT
metadata
{ "disco-role": "operating" }
description
Lightly SSL training recipes for PyTorch, Lightning, and gated distributed variants.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

training-workflows

Use this sub-skill when a task needs to turn a LightlySSL method example into a runnable training plan, choose a recipe family, adapt a training loop to local data, or debug PyTorch Lightning/distributed training behavior. It does not run long training by default; it helps future agents plan and validate safely first.

Use this sub-skill for

  • Bare PyTorch Lightly recipes that wire a backbone, projection head, transform, dataloader, loss, optimizer, and training step.
  • PyTorch Lightning module patterns, Trainer knobs, checkpoint handling, precision/device choices, and batch-size/worker decisions.
  • Adapting repo-style CIFAR/download examples to local image folders or synthetic no-download smoke checks.
  • Distributed or accelerator-aware training plans when the user explicitly asks for GPU, multi-GPU, DDP, or distributed losses.
  • Troubleshooting feature-dimension mismatches, missing checkpoints, long-running data downloads, drop_last/batch issues, and DDP hangs.

Start here

Route elsewhere

  • API component catalogs, transform/loss/head signatures, and tensor-only component debugging: use ssl-building-blocks.
  • lightly-ssl-train, lightly-embed, lightly-magic, lightly-crop, Hydra overrides, and data folder validation: use cli-data-embedding.
  • Maintainer tests, generated notebooks, docs, benchmarking utilities, and repository validation: use evaluation-maintenance.

Safe workflow order

  1. Choose the method family and recipe style: bare PyTorch, Lightning, or distributed Lightning.
  2. Validate the low-level components with ssl-building-blocks when dimensions, transforms, or losses are uncertain.
  3. Run python scripts/tiny_simclr_training_step.py --device cpu from this sub-skill before touching user data.
  4. Replace example dataset downloads with a local folder or a user-provided dataset only after the smoke passes.
  5. Scale to CUDA or DDP only after checking hardware, package backends, output paths, and runtime budget.
  6. For repository edits to examples or docs, route to evaluation-maintenance for notebook/docs/test commands.

Guardrails

  • Prefer local image folders or synthetic tensors; do not make future agents run download-heavy example scripts as the first validation step.
  • Treat distributed training as optional and explicitly gated by hardware, process group setup, and user approval.
  • Do not claim GPU or DDP verification from a CPU import; run a real backend smoke when those capabilities are required.
  • Keep CLI train/embed/magic as artifact-writing workflows. Use cli-data-embedding to build commands and set output/checkpoint paths before execution.

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