single-model-inference
Use for PaddleCV config-driven single-model classification, detection, segmentation, feature extraction, and keypoint inference workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- single-model-inference
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use for PaddleCV config-driven single-model classification, detection, segmentation, feature extraction, and keypoint inference workflows.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/configuration.md1.3 KB
- 参考文档references/README.md185 B
- 参考文档references/troubleshooting.md1.2 KB
- 参考文档references/workflows.md1.2 KB
03 SKILL.md 正文
Single-model inference
Use this sub-skill when the user wants one PaddleCV model family at a time: classification, detection, segmentation, feature extraction, or keypoint estimation.
Covers
paddlecv/configs/single_op/*.ymlPipeline(config_path=...)for a single model graphtools/predict.pystyle inference for one image, a directory of images, or a video- model catalog discovery that helps choose a single-op preset
Excludes
PaddleCV(task_name=...)system presets such as OCR, PP-Structure, ShiTu, Human, Vehicle, TinyPose, IE, SA, and TTS- custom operator registration and DAG validation
- connector/output authoring beyond the built-in single-op outputs
Read these files
../../references/task-catalog.mdfor the single-op preset list and config families.../../references/api-reference.mdforPaddleCV,Pipeline, and config parsing details.../../references/workflows.mdfor the canonical command patterns.../../references/troubleshooting.mdfor import, cache, and config errors.../../scripts/run_predict.pyfor the bundled public inference wrapper.
Typical user requests
- "Run PP-YOLOE+ on one image"
- "Use the single-op config for PP-HGNet"
- "List the supported single-op models"
- "Why does the detection config fail to resolve an input key?"
Core workflow
- Pick a
paddlecv/configs/single_op/*.ymlfile or a matching direct config path. - Run
scripts/run_predict.py --config ... --input .... - Override the YAML with
-o KEY=VALUEonly when the config needs a small local adjustment. - Use
PaddleCV.list_all_supported_models()orlist_model([...])to narrow model choice.
Model families
- Classification:
PP-LCNet,PP-LCNetV2,PP-HGNet - Detection:
PP-YOLO,PP-YOLOv2,PP-YOLOE,PP-YOLOE+,PP-PicoDet - Segmentation:
PP-HumanSegV2,PP-LiteSeg,PP-MattingV1 - Feature extraction: use embedding-style single-op configs such as
paddlecv/configs/unittest/test_feature_extraction.yml. - Keypoint: use pose-estimation single-op configs such as
paddlecv/configs/unittest/test_keypoint.yml.
Output shapes to expect
- Classification: class ids, scores, and label names.
- Detection: bounding boxes, scores, class ids, and class names.
- Segmentation: segmentation masks or alpha outputs.
- Feature extraction: bounding boxes plus features and retrieval metadata.
- Keypoint: keypoints plus keypoint scores.
Common failure modes
deviceorrun_modeis invalid.- An
Inputsentry points to the wrong prior op output. - The input is not an image, directory, video, or supported in-memory array.
- Model/download resolution fails because the cache entry is missing or the network is unavailable.
When to hand off
If the request shifts to OCR, PP-Structure, retrieval systems, or any bundled task preset, switch to system-pipelines. If the user wants to create or modify operators, switch to custom-ops.
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).