paddlex
Operate PaddleX 3.7.2 for low-code AI pipelines, module custom development, deployment, installation, and troubleshooting.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- paddlex
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Operate PaddleX 3.7.2 for low-code AI pipelines, module custom development, deployment, installation, and troubleshooting.
- disable-model-invocation
- true
02 资源文件 8
- 参考文档references/deployment-overview.md874 B
- 参考文档references/installation.md2.9 KB
- 参考文档references/module-overview.md939 B
- 参考文档references/pipeline-catalog.md854 B
- 参考文档references/repo-provenance.md2.3 KB
- 参考文档references/repo-routing-metadata.json509 B
- 参考文档references/troubleshooting.md2.8 KB
- 脚本scripts/check_paddlex_install.py2.8 KB
03 SKILL.md 正文
PaddleX operating skill
Use this skill when a task involves PaddleX, the PaddlePaddle low-code toolkit for pre-trained pipelines, model/module development, and deployment.
This skill is self-contained. Do not rely on the original PaddleX repository checkout while using it; use the bundled references and scripts here.
Route by user intent
| User intent | Read |
|---|---|
| install PaddleX, verify imports, choose extras, or fix package setup | references/installation.md, references/troubleshooting.md, scripts/check_paddlex_install.py |
| run a ready-made AI pipeline through Python or CLI | sub-skills/pipelines/ |
| export/get/edit a pipeline config, set devices/engines, save pipeline outputs | sub-skills/pipelines/ |
use create_model, validate a dataset, train, evaluate, export, predict, or convert module weights |
sub-skills/modules/ |
| deploy or accelerate a pipeline/model with HPI, serving, Paddle2ONNX, GenAI client/server, or hardware-specific runtimes | sub-skills/deployment/ |
| decide whether this skill is stale for a new checkout | references/repo-provenance.md |
Core public entry points
Installed PaddleX 3.7.2 exports these high-value APIs:
from paddlex import create_pipeline, create_predictor, create_model
from paddlex import build_dataset_checker, build_trainer, build_evaluator
Verified signatures during construction:
create_pipeline(pipeline=None, *, config=None, device=None, engine=None,
engine_config=None, pp_option=None, use_hpip=None,
hpi_config=None, **kwargs)
create_predictor(model_name, *, model_dir=None, device=None, engine=None,
engine_config=None, batch_size=1, pp_option=None,
use_hpip=False, hpi_config=None, genai_config=None, **kwargs)
create_model(model_name, model_dir=None, *args, **kwargs)
Installed CLI entry points:
paddlex --help
paddlex --pipeline image_classification --input demo.jpg --save_path output --device cpu
paddlex --get_pipeline_config image_classification
paddlex --serve --pipeline image_classification --host 0.0.0.0 --port 8080
paddlex --paddle2onnx --paddle_model_dir ./inference_model --onnx_model_dir ./onnx
paddlex_genai_server --help
Baseline install check
After installing PaddlePaddle and PaddleX, run:
python scripts/check_paddlex_install.py
A healthy baseline shows PaddleX importable, a PaddlePaddle version, CLI entry points, and whether Paddle is CPU-only or GPU-enabled.
Common decision rules
- Use pipelines for pre-trained pipeline inference and pipeline YAML orchestration.
- Use modules for custom datasets, model configs, training/evaluation/export, and
create_model. - Use deployment only after the user asks for serving, acceleration, conversion, GenAI hosting, or hardware packaging.
- CPU import success is not proof of GPU/HPI/accelerator readiness. Verify the specific PaddlePaddle wheel, plugin, and backend.
- Avoid full training, model downloads, server starts, or plugin installs until the user supplies data, runtime budget, and backend constraints.
Bundled references
references/installation.md— install order, extras, PaddlePaddle wheel choice, plugins.references/troubleshooting.md— cross-cutting package/CLI/data/backend problems.references/pipeline-catalog.md— root summary of pipeline capabilities.references/module-overview.md— root summary of module capabilities.references/deployment-overview.md— root summary of deployment capabilities.references/repo-provenance.md— source commit/evidence/staleness information.references/repo-routing-metadata.json— router metadata for managed import.
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).