onnx-and-benchmarking
Use 3DDFA_V2 ONNX Runtime acceleration and CPU latency or speed benchmark workflows.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- onnx-and-benchmarking
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use 3DDFA_V2 ONNX Runtime acceleration and CPU latency or speed benchmark workflows.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/troubleshooting.md1.2 KB
- 参考文档references/workflows.md1.1 KB
- 脚本scripts/run-latency.py910 B
- 脚本scripts/run-speed-cpu.py920 B
03 SKILL.md 正文
ONNX and benchmarking
Use this sub-skill when the task is about --onnx, ONNX conversion, CPU
latency, OMP_NUM_THREADS, latency.py, or speed_cpu.py.
When to read
Read this sub-skill when the task asks to:
- Use ONNX Runtime instead of the PyTorch model path.
- Explain or prepare
TDDFA_ONNXandFaceBoxes_ONNX. - Run end-to-end latency timing on an image.
- Run the model-only CPU microbenchmark.
- Diagnose missing
.onnxfiles, OpenMP/libomp issues, or unstable timings.
Before running
- Use
../setup-and-assets/to build and verify the baseline checkout. - Check
../../references/model-assets.mdfor which.pth,.pkl, and generated.onnxassets are involved. - Keep benchmark inputs and thread counts fixed when comparing numbers.
Latency wrapper
The latency wrapper preserves the original latency.py CLI. Put original
arguments after --:
python <skill-root>/sub-skills/onnx-and-benchmarking/scripts/run-latency.py \
--repo-root <checkout> -- \
-f <image-path> --onnx --repeated 1 --warmup true
This reports face detection, 3DMM regression, and reconstruction timing.
Speed wrapper
The speed wrapper preserves speed_cpu.py:
python <skill-root>/sub-skills/onnx-and-benchmarking/scripts/run-speed-cpu.py \
--repo-root <checkout>
It runs a random input through weights/mb1_120x120.onnx and prints the mean
and standard deviation per inference.
ONNX conversion behavior
FaceBoxes_ONNXconvertsFaceBoxesProd.pthtoFaceBoxesProd.onnxif the ONNX file is missing.TDDFA_ONNXconverts the selected.pthcheckpoint and the BFM decoder if the ONNX artifacts are missing.- Conversion is CPU-safe but mutates the checkout by writing generated
.onnxfiles.
Thread and provider notes
- The repo sets
OMP_NUM_THREADS=4in the ONNX demo paths. KMP_DUPLICATE_LIB_OK=Trueis set by the demo scripts before ONNX imports.- CPU ONNX Runtime is the baseline. CUDA/other providers are outside the required verification scope.
- On macOS, the original README notes that
libompmay be required.
Troubleshooting
Read references/troubleshooting.md for ONNX-specific issues and
../../references/troubleshooting.md for shared setup failures.
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).