AGENT SKILL

data-preparation

Routes EasyCV dataset layouts, annotation conversion, file I/O, and data-hub preparation workflows.

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

01 前言 Front Matter

name
data-preparation
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Routes EasyCV dataset layouts, annotation conversion, file I/O, and data-hub preparation workflows.
disable-model-invocation
true

02 资源文件 3

03 SKILL.md 正文

Data preparation

Use this sub-skill when the task is to prepare, validate, or convert datasets for EasyCV.

It covers dataset layout guidance, annotation conversion, OSS / local file I/O, and the repo's data-preparation helper modules.

Read these references first

  • references/data-formats.md for the expected on-disk layouts and table schemas.
  • references/workflows.md for the repo-maintained conversion helpers and dataset-specific prep flows.
  • references/troubleshooting.md for layout mismatches, missing annotation files, and OSS issues.
  • Root references/model-zoo-overview.md when the dataset choice depends on the task family.

What belongs here

Include tasks such as:

  • preparing ImageNet, CIFAR, COCO, VOC, nuScenes, Market1501, CrowdHuman, or PAI-iTAG data
  • converting annotation formats into EasyCV-ready structures
  • checking a file list, manifest, TFRecord, or table schema before training or prediction
  • using easycv.file.io for local or OSS paths
  • understanding which data shape a config expects before training

What stays elsewhere

  • Training / evaluation once the data is ready -> sub-skills/training-and-evaluation/
  • Batch inference over prepared inputs -> sub-skills/prediction-and-inference/
  • Export / optimization after training -> sub-skills/export-and-optimization/

Typical decision flow

  1. Identify the dataset family or table schema.
  2. Confirm the expected directory or file-list layout.
  3. Decide whether a conversion helper, a file I/O setup, or a manual layout fix is needed.
  4. Prepare or validate the data before touching the training command.

Common success signals

  • The dataset root matches the layout documented for the chosen family.
  • Annotation files, index files, or manifests are present where the loader expects them.
  • OSS / ODPS credentials are configured before any remote read or write.
  • The config's data roots point at the prepared dataset instead of the source notebook or download location.

Common preparation surfaces

  • easycv.file.io supports both local and OSS paths.
  • tools/prepare_data modules are packaged into the installed easycv.tools namespace.
  • Some dataset families use filelists, some use manifests, and some use TFRecords or ODPS tables.
  • The train / eval sub-skill expects the data to be ready before it starts.

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