Ground truth: numcrux transfer/donation (governance decision)

Aberta
  • Algo errado?
#987 0 comentários 0 reações 0 responsáveis Ver no GitHub

Mantenedores costumam responder em até 1 dia

Ninguém assumiu esta issue ainda.

Avaliação

Dificuldade
5/5
Tempo estimado
Mais de uma semana
Facilidade para iniciantes
25/100
Tipo de issue
Funcionalidade
Clareza
Precisa de esclarecimento
Status de atividade
Pouca atividade
Stack de tecnologia
kotlin, python, pytorch
Domínio
data, machine-learning

Direção de pesquisa

Start by reading the linked parent issue #984, then review CONTRACT.md and skainet-test-groundtruth/build.gradle.kts for the stated technical preconditions. The issue is done when the owners decide whether to transfer the corpus, what happens to gradienttracer and licensing, and whether CI should use a numcrux-hosted corpus; it does not specify an implementation task.

Escrita pelo modelo de indexação a partir do texto da issue.

Descrição

question sub-issue

Sub-issue of #984.

skainet-ground-truth's actual Python content is already framework-agnostic — it
imports only torch and gradienttracer, zero SKaiNET/Kotlin coupling. The "SKaiNET"
branding is only in the repo name and README framing, not the code, matching
numcrux's own stated mission ("a public, neutral,
versioned reference-data commons... framework-agnostic and language-agnostic... will
start with the existing numerical test suite from SKaiNET").

skainet-test-groundtruth/build.gradle.kts's fixture-source path is now overridable
(-PgroundTruthSourceDir, defaults to the sibling-checkout convenience it always had),
and CONTRACT.md documents the GGUF tensor-naming/metadata conventions explicitly — so
SKaiNET's consumer no longer needs to change if/when the underlying fixtures move to a
numcrux-owned repo. That was the technical precondition; this issue is the remaining
non-technical one.

Open questions (for whoever owns this decision)

  • Does skainet-ground-truth's content get donated/transferred to the numcrux org
    outright, or does numcrux start a fresh corpus and skainet-ground-truth stays as
    SKaiNET's own (possibly narrower, possibly diverging) fixture set?
  • If transferred: does gradienttracer (the execution/serialization engine) move too,
    or stay under SKaiNET-developers as infrastructure numcrux depends on?
  • License/attribution framing for a neutral commons vs. a SKaiNET-branded repo.
  • Whether SKaiNET's CI (.github/workflows/ground-truth.yml) should point at the
    numcrux-hosted corpus once one exists, via -PgroundTruthSourceDir, or continue
    using skainet-ground-truth as a curated subset.

Not blocking anything today — filed so the decision has a home when it's ready to be
made, and so the "near future" framing from the original ask has a concrete place to
land.

Linguagem predominante
Kotlin
Estrelas
52
Forks
15
Merge médio
23h 45min
PRs com merge (30d)
80

Preparar o ambiente

Primeiros passos

  1. Leia a issue inteira e depois o guia de contribuição do projeto.
  2. Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
  3. Faça um fork do repositório e trabalhe em uma branch.
  4. Abra um pull request que referencie o número da issue.

Mais de SKaiNET-developers/SKaiNET

Todas as issues de SKaiNET-developers/SKaiNET

Issues semelhantes

Mais issues de Kotlin

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.