AGENT SKILL

training

Guides CUDA-only SiamMask base/refine and SiamRPN training, checkpoint resume, experiment configs, and training setup validation.

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

01 前言 Front Matter

name
training
license
MIT
metadata
{ "disco-role": "operating" }
description
Guides CUDA-only SiamMask base/refine and SiamRPN training, checkpoint resume, experiment configs, and training setup validation.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

SiamMask Training

Use this sub-skill when the task is to train or resume SiamMask/SiamRPN models, inspect experiment configs, plan data requirements, or debug CUDA/batch/checkpoint issues.

Hard Requirement

The native training scripts call CUDA APIs unconditionally (model.cuda() and DataParallel(...).cuda()). Treat CUDA as required with no CPU substitute for real training. CPU-only work can validate configs and commands, but it does not verify training runtime.

Prerequisites

  • Read root install/setup and model overview.
  • Run the root environment probe with --expect-cuda yes before real training.
  • Prepare training data through ../data-preparation/SKILL.md. The configs expect crop511 directories and JSON indexes.
  • Obtain resnet.model or an appropriate base checkpoint before training from scratch/refine.

Main Routes

User intent What to do
Train SiamMask base Read references/workflows.md and use scripts/run_training.py base in dry-run mode first.
Train the refine/sharp model Read references/workflows.md; provide --pretrained <base-checkpoint>.
Train the SiamRPN baseline Read references/workflows.md and use scripts/run_training.py siamrpn.
Inspect config/data paths Read references/config-reference.md; the bundled runner summarizes dataset roots and JSON paths before launching.
Debug training failure Read references/troubleshooting.md.

Bundled Helper

Use scripts/run_training.py to compose commands and inspect config paths. It:

  • Defaults to dry-run and prints the command, cwd, PYTHONPATH, and config dataset summary.
  • Sets PYTHONPATH to include checkout root and experiment directory.
  • Supports base, refine, and siamrpn modes.
  • Lets you set --gpu <CUDA_VISIBLE_DEVICES> before execution.
  • Only starts training when --run is supplied.

Example dry-run:

python scripts/run_training.py --repo-root <siammask-checkout> --gpu 0,1,2,3 base \
  --config config.json --batch 64 --workers 20 --epochs 20

Add --run before base only after the user approves a long CUDA training job.

Do Not

  • Do not claim training is verified from CPU imports or command help.
  • Do not start full training without checking disk space, dataset availability, checkpoint paths, and user approval.
  • Do not reuse refine checkpoints with base-only configs or box-only SiamRPN flags.
  • Do not hide missing dataset roots; training scripts will otherwise fail after allocating GPU/data-loader resources.

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