AGENT SKILL

data-and-checkpoints

Use this repo skill for Swin-Transformer ImageNet folder, zip, ImageNet-22K, SimMIM data layouts, and checkpoint resume/pretrained loading behavior.

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

01 前言 Front Matter

name
data-and-checkpoints
license
MIT
metadata
{ "disco-role": "operating" }
description
Use this repo skill for Swin-Transformer ImageNet folder, zip, ImageNet-22K, SimMIM data layouts, and checkpoint resume/pretrained loading behavior.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

data-and-checkpoints

Use this sub-skill when a task involves data roots, zipped ImageNet maps, ImageNet-22K annotations, SimMIM input batches, checkpoint files, pretrained-vs-resume semantics, or class-head remapping.

What this sub-skill covers

  • ImageNet-1K folder layout for torchvision.datasets.ImageFolder.
  • Zipped ImageNet layout with train.zip, val.zip, train_map.txt, and val_map.txt.
  • ImageNet-22K layout and JSON map expectations for IN22KDATASET.
  • SimMIM pretraining data shape: images plus generated masks.
  • Checkpoint dictionary expectations for load_checkpoint and load_pretrained.
  • ImageNet-22K-to-1K classifier head remapping through the 22K class map.

Boundaries

  • For model constructors and output tensor shapes, use core-models.
  • For supervised train/eval commands, use training-eval-cli.
  • For SimMIM launch commands and mask-loss details, use simmim-workflows.
  • For MoE checkpoint sharding and Tutel-specific behavior, use moe-and-acceleration.

Standard workflow

  1. Identify the data mode: imagenet folder, zipped imagenet, imagenet22K, or SimMIM pretrain/fine-tune.
  2. Read references/data-formats.md for the exact structure.
  3. Run scripts/validate_imagenet_layout.py on a small or full data root before composing a training command.
  4. If a checkpoint is involved, read references/checkpoint-loading.md to choose --resume vs --pretrained and anticipate classifier-head behavior.
  5. Use references/troubleshooting.md if data loaders report empty datasets, malformed maps, or checkpoint key mismatches.

Safe checks

The bundled validator checks schema and small samples only. It does not download ImageNet, rewrite validation folders, unzip archives, or load model weights.

Linked files

  • references/data-formats.md - folder, zip, 22K, and SimMIM data structures.
  • references/checkpoint-loading.md - resume/pretrained/checkpoint remap behavior.
  • references/troubleshooting.md - common data and checkpoint failures.
  • scripts/validate_imagenet_layout.py - safe schema validator for folder, zip, and ImageNet-22K JSON map layouts.

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