Przejdź do treści

Training Job v1

Source schema: doc/schemas/training-job.v1.schema.json

Machine-readable schema for adapter-first specialization jobs built from approved corpus entries.

Governing Basis

Project Lineage

Requirements

Stories

Fields

Field Required Shape Description
schema/v yes const: 1 Schema version.
job/id yes string Stable identifier of the specialization job.
base-model/ref yes string Reference to the immutable base model on which specialization is built.
method yes enum: lora, qlora Adapter-first specialization method.
dataset/refs yes array Corpus or dataset references used by the job.
policy/profile yes string Training policy profile applied to this job.
started-at yes string Job start timestamp.
ended-at no string Job completion timestamp.
operator/ref yes string Human or policy actor responsible for starting the job.
status no enum: running, completed, failed, canceled Current or terminal job status.
eval-report/ref no string Reference to evaluation output when the job completes.
creator/refs no array Creator or contributor references that should survive into attribution-sensitive adapter artifacts.
policy_annotations no object Optional implementation-local annotations that do not change the core training-job semantics.

Conditional Rules

Rule 1

When:

{
  "properties": {
    "status": {
      "const": "completed"
    }
  },
  "required": [
    "status"
  ]
}

Then:

{
  "required": [
    "ended-at",
    "eval-report/ref"
  ]
}

Rule 2

When:

{
  "properties": {
    "ended-at": {
      "type": "string"
    }
  },
  "required": [
    "ended-at"
  ]
}

Then:

{
  "required": [
    "status"
  ]
}

Field Semantics

schema/v

  • Required: yes
  • Shape: const: 1

Schema version.

job/id

  • Required: yes
  • Shape: string

Stable identifier of the specialization job.

base-model/ref

  • Required: yes
  • Shape: string

Reference to the immutable base model on which specialization is built.

method

  • Required: yes
  • Shape: enum: lora, qlora

Adapter-first specialization method.

dataset/refs

  • Required: yes
  • Shape: array

Corpus or dataset references used by the job.

policy/profile

  • Required: yes
  • Shape: string

Training policy profile applied to this job.

started-at

  • Required: yes
  • Shape: string

Job start timestamp.

ended-at

  • Required: no
  • Shape: string

Job completion timestamp.

operator/ref

  • Required: yes
  • Shape: string

Human or policy actor responsible for starting the job.

status

  • Required: no
  • Shape: enum: running, completed, failed, canceled

Current or terminal job status.

eval-report/ref

  • Required: no
  • Shape: string

Reference to evaluation output when the job completes.

creator/refs

  • Required: no
  • Shape: array

Creator or contributor references that should survive into attribution-sensitive adapter artifacts.

policy_annotations

  • Required: no
  • Shape: object

Optional implementation-local annotations that do not change the core training-job semantics.