pfllib
Routes PFLlib tasks for dataset preparation, federated-learning experiments, and library extension.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- pfllib
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Routes PFLlib tasks for dataset preparation, federated-learning experiments, and library extension.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/repo-provenance.md1.4 KB
- 参考文档references/repo-routing-metadata.json547 B
- 参考文档references/troubleshooting.md2.6 KB
- 脚本scripts/check_install.py3.5 KB
- 脚本scripts/scan_registry.py3.7 KB
03 SKILL.md 正文
PFLlib
Use this skill for PFLlib dataset generation, experiment execution, and code extension tasks.
Start here
- Create or verify a Python 3.11 environment with the PFLlib runtime stack: PyTorch, torchvision, torchtext, cvxpy, NumPy < 2, and the scientific stack used by the repository.
- Run
scripts/check_install.py --repo-root <checkout>to confirm imports and CUDA availability. - Run
scripts/scan_registry.py --repo-root <checkout>when you need a quick view of the supported dataset, model, and algorithm surface. - Pick the sub-skill that matches the task family below.
Routes
data-preparation
Use this route to generate or validate client-split datasets such as MNIST,
AG News, Amazon Review, HAR, and the other built-in scenario generators.
Read sub-skills/data-preparation/SKILL.md first.
experiments
Use this route to launch FL runs, tune CLI flags, inspect results, or check
privacy/system-condition settings. Read sub-skills/experiments/SKILL.md
first.
extension
Use this route when you need to add a new algorithm, dataset, model, or
optimizer to the repo. Read sub-skills/extension/SKILL.md first.
Shared helpers
scripts/check_install.py— run this first when the environment or CUDA stack is in doubt.scripts/scan_registry.py— use this to inspect the supported registry before editing or launching.
Read these references
references/repo-provenance.mdwhen you need to compare the skill against a checkout or decide whether it should be refreshed.references/troubleshooting.mdfor cross-cutting install, import, CUDA, cvxpy, and working-directory problems.references/repo-routing-metadata.jsonfor the router placement data used by the live repo-skills router.
Notes
- This repository is script-driven rather than a packaged library; use the bundled check and launch helpers against a local checkout.
- If a dataset tree is missing, route to
data-preparationbefore launching an experiment. - If the task is to change the codebase, not just run it, route to
extension.
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).