メニュー
Layer 2 (Tuner) specialist of haipipe-nn: defines the hyperparameter search space and the tuner that drives sweeps. Called by /haipipe-nn; direct invocation works layer-scoped.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/jluo41/Tools --skill haipipe-nn-tunerコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
PostFn specialist -- designs/reviews the response-formatting function in an Endpoint_Set. One of 5 inference Fn-types. Called by /haipipe-end when intent references PostFn, response formatting, post-processing, or `post`.
Run any Stage 6 endpoint work: routes intent to the right specialist across Fn-type, artifact-verb, develop-target, and deploy-target axes. Use for designing inference Fns, packaging Endpoint_Sets, training a build, local inference tests, or deploying anywhere. Trigger: endpoint, deploy, develop, train, package, inference Fn, MetaFn, TrigFn, PostFn, Src2InputFn, Input2SrcFn, /haipipe-end.
Src2InputFn specialist -- designs/reviews the record->wire-payload function in an Endpoint_Set (serializes a ProcessedDF record into JSON the model can ingest). Platform-specific: one impl per deploy platform (SageMaker flat JSON vs Databricks dataframe_records); --platform picks (default sagemaker). Called by /haipipe-end when intent references Src2InputFn, record-to-payload serialization, or `src2input`.
SKILL.md を表示中
SOC 職業分類に基づく
| name | haipipe-nn-tuner |
| description | Layer 2 (Tuner) specialist of haipipe-nn: defines the hyperparameter search space and the tuner that drives sweeps. Called by /haipipe-nn; direct invocation works layer-scoped. |
| argument-hint | [function] [args...] |
| allowed-tools | Bash, Read, Write, Edit, Grep, Glob |
| metadata | {"version":"0.1.1","last_updated":"2026-07-04","summary":"Layer 2 (Tuner) specialist of haipipe-nn."} |
Layer 2 specialist.
Owns the tuner classes in code/hainn/tuner/ — search space declarations, hyperparameter sampling, and the loop that wraps L1 algorithms.
Called by the /haipipe-nn orchestrator; can also be invoked directly.
Function axis: dashboard | review | generate | test
/haipipe-nn-tuner -> dashboard: registered tuners
/haipipe-nn-tuner dashboard -> same
/haipipe-nn-tuner review [class_name] -> structural review of a tuner class
/haipipe-nn-tuner generate -> scaffold a new tuner / search space
/haipipe-nn-tuner test [class_name] -> run a short tuner smoke sweep
Invocation This skill's ref Umbrella's fn doc
------------- ---------------------- ------------------------------
dashboard ref/concepts.md ../haipipe-nn/fn/fn-dashboard.md
review ref/concepts.md ../haipipe-nn/fn/fn-review.md
generate ref/concepts.md +
../haipipe-nn-instance/
ref/concepts.md ../haipipe-nn/fn/fn-generate.md
test ref/concepts.md ../haipipe-nn/fn/fn-test.md
(no fn arg) ref/concepts.md (ref-only mode)
generate reads the L3 (instance) ref because the tuner's output (best-config + ckpt) becomes the input contract for ModelInstance.
Step 0: Read ../haipipe-nn/ref/overview.md.
Mandatory.
Step 1: Parse args.
Function vocabulary: dashboard | review | generate | test.
Step 2: Read this skill's ref/concepts.md for L2 specifics.
Step 3: Read the umbrella fn doc.
Step 4: For generate, also read ../haipipe-nn-instance/ref/concepts.md.
Step 5: Execute, scoped to L2 (tuner only — no algorithm class edits).
Step 6: Emit the structured tail.
Owns:
code/hainn/tuner/Upstream dependency (L1):
Tuner search space must match the algorithm's __init__ signature. If a
param is missing or extra, escalate to /haipipe-nn-algo review.
Hand-off contract (L2 -> L3):
ModelInstance creates this Tuner via the model_tuner registry and calls
fit() / save_model(key, model_dir); the Tuner owns training and
serialization inside the instance's model_dir. Verify against
../haipipe-nn-instance/ref/concepts.md.