sweeps-and-launch
Define and run W&B hyperparameter sweeps and W&B Launch jobs, queues, agents, and launch sweeps across local process, local container, Kubernetes, SageMaker, and Vertex resources.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- sweeps-and-launch
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Define and run W&B hyperparameter sweeps and W&B Launch jobs, queues, agents, and launch sweeps across local process, local container, Kubernetes, SageMaker, and Vertex resources.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/launch.md8.2 KB
- 参考文档references/sweeps.md5.4 KB
- 参考文档references/troubleshooting.md5.5 KB
- 脚本scripts/validate_sweep_config.py7.9 KB
03 SKILL.md 正文
W&B Sweeps and Launch
Use this sub-skill when an agent needs to create sweep configs, register sweeps with wandb.sweep, run bounded sweep agents with wandb.agent or wandb agent, or submit W&B Launch jobs/queues/agents. Use other W&B sub-skills for basic wandb.init tracking, artifact/model registry I/O, or global CLI setup.
Route by task
- Sweep definition and execution: Read references/sweeps.md for config shape,
wandb.sweep,wandb.agent,wandb sweep, andwandb agentpatterns. - Launch jobs and queues: Read references/launch.md for
wandb launch,wandb job,wandb launch-agent,wandb launch-sweep, resource selection, and queue handoff patterns. - Failure repair: Read references/troubleshooting.md when configs are rejected, agents run forever, child processes ignore termination, Launch extras/backends are missing, or queue/resource/credential assumptions are unsafe.
- Offline config linting: Run
python scripts/validate_sweep_config.py path/to/sweep.yamlbefore creating a sweep when you need a safe local check for the required sweep keys and parameter shape.
Safe defaults
- Prefer bounded sweep agents while iterating: pass
counttowandb.agent(...)or--counttowandb agentunless the user explicitly wants continuous workers. - Use exactly one execution style per sweep: either a
programin the sweep config for CLI agents or a Pythonfunctionpassed towandb.agent; do not require both. - For Launch queues, choose
--resourcedeliberately when the queue has no resource configuration:local-process,local-container,kubernetes,sagemaker, orgcp-vertex. - Treat cloud Launch examples as templates only. Do not invent credentials, IAM roles, Kubernetes contexts, container registries, S3/GCS buckets, or Vertex project/region values.
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).