extension
Routes PFLlib code changes that add or update algorithms, models, optimizers, and dataset generators.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- extension
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Routes PFLlib code changes that add or update algorithms, models, optimizers, and dataset generators.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/api-reference.md3.3 KB
- 参考文档references/troubleshooting.md2.3 KB
- 参考文档references/workflows.md1.9 KB
03 SKILL.md 正文
Extension
Use this route when you need to change PFLlib itself rather than run an existing benchmark.
Use this when
- You want to add a new federated-learning algorithm, client, or server.
- You want to add a new dataset generator or preprocessing flow.
- You want to add a new model, optimizer, or registry entry.
- You need to verify that a new code path is wired into
main.pycorrectly.
Read these references
references/api-reference.mdfor the Server/Client hooks and registry expectations.references/workflows.mdfor the step-by-step add-a-feature checklists.references/troubleshooting.mdfor registration, shape, and dependency mistakes.
Run these helpers
scripts/scan_registry.pyfrom the root route to inspect the current dataset, model, and algorithm registry before you edit it.scripts/check_install.pyfrom the root route if the new feature introduces a dependency likecvxpy,torchtext, ortorchvision.
What belongs here
Include code changes and documentation for:
- new
server*.py/client*.pypairs - new dataset generators under
dataset/ - new model classes under the model registry
- new optimizer hooks
- registry edits in
system/main.py - any support code required to make the new feature selectable from the CLI
What does not belong here
- Launching an already-registered experiment; route that to
experiments. - Generating or validating client-split datasets that already exist; route that
to
data-preparation. - Long result-analysis tasks after the change is already wired; route that back
to
experiments.
Common workflow
- Inspect the current registry and the relevant base classes.
- Implement the new code path.
- Wire it into
main.pyor the dataset loader as needed. - Re-check the registry and run a tiny smoke path from the experiments route.
- Only then consider larger validation or downstream benchmarking.
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).