export-and-speed
Routes Ultra-Fast-Lane-Detection TorchScript export, synthetic speed checks, and deployment benchmarking workflows.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- export-and-speed
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Routes Ultra-Fast-Lane-Detection TorchScript export, synthetic speed checks, and deployment benchmarking workflows.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/api-reference.md1.2 KB
- 参考文档references/export-workflows.md1.5 KB
- 参考文档references/speed-and-deployment.md1.2 KB
- 参考文档references/troubleshooting.md2.2 KB
- 脚本scripts/benchmark_synthetic.py2.6 KB
- 脚本scripts/export_torchscript.py3.0 KB
03 SKILL.md 正文
export-and-speed
Use this sub-skill when a task is about exporting a lane model for deployment, timing its throughput, or understanding the C++/LibTorch demo path.
Read this when
- You need to export a checkpoint to TorchScript.
- You need a quick synthetic speed benchmark.
- You need to explain the C++/LibTorch deployment example.
- You need to reason about
cls_dim, anchors, device choice, or half precision for export and timing.
What this sub-skill owns
- TorchScript export commands and model input shape.
- Synthetic throughput timing.
- Deployment caveats for the C++ demo.
- Speed-related checkpoint, device, and dimension assumptions.
What it does not own
- Training command construction: see
training. - Dataset layout and conversion: see
data-and-config. - Accuracy evaluation and demo metrics: see
evaluation-and-visualization.
Start here
- Read
references/export-workflows.mdfor the export and benchmark flow. - Read
references/speed-and-deployment.mdfor the LibTorch/OpenCV notes and practical timing caveats. - Read
references/api-reference.mdfor the verified model signature and dimension assumptions. - Read
references/troubleshooting.mdfor hardcoded path, checkpoint, and device issues. - Run
scripts/export_torchscript.pywhen you need a portable TorchScript file. - Run
scripts/benchmark_synthetic.pywhen you need a configurable synthetic timing check.
Typical flow
- Decide whether the user wants a CPU or CUDA export/benchmark path.
- Confirm the checkpoint file and model dimensions.
- Export or benchmark using a helper that accepts explicit paths and device options.
- If the user wants C++ deployment, read the deployment caveats before building anything.
Caution points
- The source
export.pyhas a hardcoded checkpoint path and CUDA map_location. - The source speed scripts are demos, not robust command-line tools.
- The C++ example hardcodes LibTorch and OpenCV paths.
cls_dimmust match the chosen dataset family and backbone setup.
Reference and script links
references/export-workflows.mdfor export patterns and safe invocation advice.references/speed-and-deployment.mdfor deployment and benchmark caveats.references/api-reference.mdfor the verified signature notes.references/troubleshooting.mdfor export and timing failures.scripts/export_torchscript.pyfor a parameterized export helper.scripts/benchmark_synthetic.pyfor a configurable speed helper.
04 同仓库更多 Skills
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.
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.
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.
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.
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.
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).