AGENT SKILL

office-benchmarks

Routes LibMTL's Office-31 and Office-Home multi-input classification workflows.

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

01 前言 Front Matter

name
office-benchmarks
license
MIT
metadata
{ "disco-role": "operating" }
description
Routes LibMTL's Office-31 and Office-Home multi-input classification workflows.
disable-model-invocation
true

02 资源文件 56

03 SKILL.md 正文

office-benchmarks

Use this sub-skill for the Office-31 and Office-Home benchmark workflows.

Covers

  • Office-31 multi-input classification.
  • Office-Home multi-input classification.
  • The bundled office split layout and dataloader recipe.
  • A self-contained Office runtime package under scripts/office_runtime/.
  • An unpackaged launcher at scripts/main.py plus the install entry point scripts/install_office_runtime.py.
  • ResNet-18 based encoders and task-specific linear decoders.

Does not cover

  • NYUv2 or Cityscapes dense prediction.
  • QM9 graph regression.
  • PAWS-X multilingual text classification.
  • Core API or custom method development unless the question is specifically about the office benchmark wiring.

When to use this sub-skill

Choose this route when the user asks things like:

  • "How do I train Office-31?"
  • "What is the Office-Home layout?"
  • "Why must multi_input be true here?"
  • "How do the office dataloaders map tasks to domains?"
  • "How do I validate the split files?"

Read next

  • ../../references/configuration.md for the shared flags and method kwargs.
  • ../../references/troubleshooting.md for cross-cutting install and runtime failures.
  • references/workflows.md for the command pattern.
  • references/task-contracts.md for the task dictionary, loader, and encoder/decoder shapes.
  • references/data-layouts.md for the expected split and image trees.
  • references/troubleshooting.md for office-specific failures.

Workflow

  1. Confirm the dataset family: office-31 or office-home.
  2. Confirm that the user expects multi-input classification.
  3. Validate the split text files and the image root.
  4. Run scripts/check_office_data.py to confirm the bundled split files and image root before training.
  5. Install the self-contained runtime package with scripts/install_office_runtime.py when you need the console entry point.
  6. Launch the benchmark through scripts/main.py or the installed libmtl-office entry point, not through an external checkout.
  7. Allow the pretrained resnet18 backbone to download or use cache.

Critical constraints

  • multi_input must be True.
  • The bundled runtime builds one dataloader per domain/task from its own split files.
  • The task list depends on the dataset family:
    • Office-31: amazon, dslr, webcam
    • Office-Home: Art, Clipart, Product, Real_World

Exit criteria

Leave this sub-skill when the user has the correct dataset family, split files, run command, and failure-recovery notes for the office benchmark.

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