Hacktoberfest 2026:メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

[SKEEP-006]: Input front ends and model identity

オープン
#1,331 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
5/5
見積もり時間
1週間以上
初心者へのやさしさ
15/100
issue の種類
機能追加
明瞭さ
おおむね明確
活発さ
活発
技術スタック
huggingface, kotlin

調査の方向性

Read the draft proposal at docs/modules/skeep/pages/006-input-frontends-and-model-identity.adoc and its registration in docs/modules/skeep/nav.adoc. The proposal is still a draft, and implementation is already underway in a fork; wait for the maintainer to accept it and review the linked implementation work before taking on any sub-issue. Done means the proposal’s status and follow-up issues reflect the agreed scope.

索引モデルが issue の本文から書いたものです。

説明

skeep tracking

Proposal document: docs/modules/skeep/pages/006-input-frontends-and-model-identity.adoc — prepared in a personal fork: https://github.com/michalharakal/SKaiNET-core-upstream/blob/feature/skeep-006-input-frontends/docs/modules/skeep/pages/006-input-frontends-and-model-identity.adoc
Status: Draft
Branch: feature/skeep-006-input-frontends (fork; the upstream pull request follows once the #1321 good-first-issue lanes have landed, so the contributors working them are not disturbed)

Trigger

Public Kotlin API (new modules skainet-model-identity, skainet-frontend-core, skainet-frontend-text; the sk.ainet.io.tokenizer package becomes deprecated aliases) · compatibility/migration policy (three-release deprecation of the superseded identity switches and of the SKaiNET-transformers legacy tokenizer stack) · runtime integration (the resolver is what model loaders, Java callers and cartridge facades use).

Summary

A checkpoint is weights, an identity and the recipe that turns raw input into the tensors the weights expect. Today the recipe is re-derived downstream: SKaiNET-transformers carries a second tokenizer stack that disagrees with the engine's, model identity is split three ways in the engine and a fourth way downstream, and audio/image preprocessing has no home at all. The proposal introduces FrontEnd<I, O> with Tokenizer as its first implementor, a string-based ModelIdentity with an immutable ModelFamilyRegistry that weight mapping and front ends both key on, and FrontEndResolver as the single entry point. Algorithms live in the engine, golden-tested against the Hugging Face tokenizers library; families (downstream) contribute policy only. Audio and image features follow on the same contract in later lanes.

Related DARC features

  • #1321 — tokenizer.json parity for BPE encoders (the lane fixes #1323–#1326 are folded into the moved implementations; the lanes stay open for their contributors and are closed by the upstream PR of this proposal or by their own PRs, whichever lands first)
  • SKaiNET-developers/SKaiNET-transformers#471 — legacy tokenizer defects and parity gate (its lanes become the migration gate)
  • SKaiNET-developers/SKaiNET-transformers#330 — two GGUF tokenizer paths disagree (answered by ModelFamilyDescriptor + FrontEndResolver)

Sub-issues

Filed once the proposal is Accepted: engine F2 (contract, identity, resolver — implemented in the fork branch), F4 log-mel front end, F5 image front ends; SKaiNET-transformers F3 (consume engine tokenizers — implemented in the fork branch feature/skeep-006-frontend-consumer of https://github.com/michalharakal/SKaiNET-transformers), F6 audio families.

Status upkeep

  • Proposal registered in docs/modules/skeep/nav.adoc and the "Current Proposals" table (fork branch)
  • Maintainer moved status to Accepted
  • Implementation PR(s) linked here and flipped the proposal's Status: to Implemented
主要言語
Kotlin
スター
52
フォーク
15
平均マージ
1日 4時間
マージ済み PR(30日)
37

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

SKaiNET-developers/SKaiNET のほかの issue

SKaiNET-developers/SKaiNET の issue をすべて見る

似ている issue

Kotlin の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。