AGENT SKILL

model-sharing

Prepare, inspect, and safely smoke-test Asteroid publishable models and upload metadata.

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

01 前言 Front Matter

name
model-sharing
license
MIT
metadata
{ "disco-role": "operating" }
description
Prepare, inspect, and safely smoke-test Asteroid publishable models and upload metadata.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

Model sharing and publishing

Use this sub-skill when the user wants to save a publishable Asteroid model, register an old sample rate, or prepare metadata for Zenodo upload.

Typical triggers

  • save_publishable(...)
  • upload_publishable(...)
  • asteroid-upload
  • asteroid-register-sr
  • Zenodo metadata, model cards, or publishable artifacts
  • legacy checkpoints missing sample_rate

What to do first

  1. Confirm whether the user wants only a local publishable artifact or an actual upload.
  2. Check whether the checkpoint already contains:
    • model_args
    • state_dict
    • dataset
    • licenses
    • infos
  3. Determine whether a token or sandbox upload is involved.

Standard workflow

  • Read references/publishable-models.md for the local artifact contract.
  • Read references/cli-reference.md for the public upload and sample-rate registration entrypoints.
  • Read references/troubleshooting.md when the upload path is blocked by credentials or network dependencies.
  • Use scripts/smoke_publishable.py for a safe local save/metadata smoke check.

Safe boundaries

  • Local save_publishable(...) smoke checks are fine.
  • Real Zenodo uploads are credential-bound and should stay gated.
  • Do not treat network upload as a default verification step.
  • asteroid-register-sr is for legacy checkpoints that were saved before sample_rate was present.

Useful inputs

  • uploader name
  • affiliation
  • git username
  • Zenodo token or sandbox token
  • recipe name
  • final metrics dict
  • dataset/task/license metadata

Troubleshooting reminders

  • The model dict must include the required keys before saving or uploading.
  • upload_publishable(...) needs a token unless one is present in ACCESS_TOKEN.
  • unit_test=True is useful for safe metadata inspection, not a real upload.

Inputs to inspect

  • whether the user wants local artifact prep or a real publication step
  • uploader name, affiliation, and git username
  • whether a Zenodo token or sandbox token is available
  • whether the checkpoint is a fresh model or a legacy checkpoint missing sample_rate

Smoke sequence

  1. Build a tiny model.
  2. Serialize it.
  3. Attach dataset and license metadata.
  4. Save a publishable artifact in a temporary directory.
  5. Inspect the generated model.pth without publishing it.

What to avoid

  • Do not make real publication the default path.
  • Do not pretend that a missing token is a transient coding error.
  • Do not claim the sample-rate registration helper is for modern checkpoints.
  • Do not recommend network uploads when the user only needs a local smoke.

Metadata reminders

  • Required publishable keys must already be present in the model dict.
  • The license list should come from the dataset or training source.
  • The final metrics dict should be concrete, even for a smoke path.
  • Recipe name and training config are part of the published metadata story.

Good questions to ask when unclear

  • Is this just a local publishable artifact or an actual upload?
  • Do you have a token available?
  • Is the checkpoint legacy or modern?
  • What dataset/task name should appear in the metadata?

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