AGENT SKILL

deployment-and-projects

Operate FastReID deployment export surfaces, optional ONNX/Caffe/TensorRT dependencies, FastRT, and extension-project registration patterns.

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

01 前言 Front Matter

name
deployment-and-projects
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Operate FastReID deployment export surfaces, optional ONNX/Caffe/TensorRT dependencies, FastRT, and extension-project registration patterns.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

FastReID deployment and projects

Use this sub-skill when a FastReID task involves model export, deployment runtime checks, ONNX Runtime/Caffe/TensorRT inference, FastRT C++ TensorRT use, or extension projects such as FastAttr, FastClas, FastDistill, FastFace, FastRetri, FastTune, PartialReID, and NAIC20.

Route boundaries

  • Core model builders, tensor shapes, DefaultPredictor-style feature extraction, and checkpoint loading details belong to ../modeling-and-inference/.
  • Shared dataset registry mechanics, built-in dataset layouts, and custom dataset classes belong to ../data-and-datasets/.
  • Standard training/evaluation launch, distributed flags, resume, checkpoints, and evaluator output belong to ../training-and-evaluation/.
  • Source-only package setup, config merge, _BASE_, and model-zoo recipe selection belong to ../setup-and-configuration/.

Bundled references

  • references/deployment-workflows.md — use for ONNX, ONNX Runtime, Caffe, TensorRT, and FastRT prerequisites, export/inference interface templates, validation comparisons, and backend limitations.
  • references/project-extensions.md — use for the extension-project map, import-before-config rules, registry/config injection patterns, and project-specific dependency notes.
  • references/troubleshooting.md — use when parser help fails before argument parsing, optional deployment packages are missing, exported models disagree, TensorRT engines mismatch devices, Caffe protobuf/helpers are absent, or project imports/configs fail.

Bundled scripts

  • scripts/check_deployment_dependencies.py — safe optional-dependency probe for onnx, onnxruntime, onnxoptimizer, onnxsim, caffe, tensorrt, cv2, and torch; use it before export/runtime work or when a deployment entrypoint fails at import time.
  • scripts/project_import_probe.py — safe import probe for selected FastReID extension projects; pass an explicit FastReID application checkout with --repo-root and one or more --project values to classify registration/import failures without training.

Operating facts

  • ONNX, ONNX Runtime, Caffe, TensorRT, PyCUDA, Ray Tune, bcolz, and mxnet are optional stacks, not part of the minimum FastReID CPU inspection environment.
  • ONNX export code imports onnx, onnxoptimizer, and onnxsim before argument parsing in this FastReID version; a missing onnx package can make --help fail before showing parser output.
  • TensorRT export code imports tensorrt before argument parsing; a missing TensorRT Python package can make --help fail before showing parser output.
  • TensorRT/Caffe/FastRT workflows are hardware/runtime-gated and should not be claimed verified unless their native runtimes, drivers, model weights, and target devices are available.
  • Project packages must usually be imported before merging or building configs that name their datasets, meta-architectures, heads, backbones, or project-only config keys.

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