Skills 2026-08-15 17:52 UTC
dot-skillsv0.1.0
Version [v0.1.0] committed by pproenca. Update Category: Feature Addition & Enhancement. Commit details: Add ray + ray-llm skill kit (v0.1.0). This update introduces new capabilities and module extensions.
CommitRaw commit message
Add ray + ray-llm skill kit (v0.1.0) ray: library-reference distillation for production open-source Ray, pinned to 2.57.0 — 26 rules / 6 categories (train, serve, data, core, tune, prod) correcting older-corpus defaults: Train V2 as default, ray.air session reporting, Trainer-inside-Tuner, map_batches concurrency= (the twice-reversed parameter), DatasetPipeline/to_torch, max_concurrent_queries, RayServeHandle, ray.state, plus KubeRay production topology (gcsFaultToleranceOptions, RayJob/RayService, Jobs API, baked images). Verified against the unpacked wheel; 13 idiom groups executed live on a local 2.57.0 runtime. ray-llm: sibling rule pack for LLM workloads — ray.serve.llm (LLMConfig + build_openai_app, OpenAiIngress casing trap, vLLM version pin) and ray.data.llm (build_processor rename, stage configs, CPU-default accelerator_type), placement (bundle_per_worker trap), autoscaling/routing (PrefixCacheAffinityRouter), dynamic LoRA multiplexing. 13 rules / 5 categories; every example config constructed under CPU-only pydantic validation, engine behavior source-verified. Both reviewed by skill-reviewer with fact-check briefs; all blockers and should-fixes applied (reviewers caught 5 example-level bugs by executing configs and sweeps). Claude-Session: https://claude.ai/code/session_013dyPT33B2MhEjhbvUrecBZ