Entrenamiento

Entrena un modelo personalizado con tu dataset. El entrenamiento es asincrono y puedes monitorear el progreso.

Iniciar entrenamiento

bash

PROJ_ID="proj_1a2b3c4d"
DATASET_ID="dataset_abc123"

curl -X POST https://api.astrapp.lat/api/v1/projects/$PROJ_ID/training \
  -H "Authorization: Bearer sk_live_xxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Modelo v1",
    "dataset_id": "'$DATASET_ID'",
    "base_model": "gpt-4",
    "parameters": {
      "learning_rate": 0.001,
      "epochs": 10,
      "batch_size": 32
    }
  }'

Parametros de entrenamiento

learning_rate0.001Tasa de aprendizaje (0.0001-0.01)

epochs10Numero de epocas (5-50)

batch_size32Tamano del batch (8-128)

early_stoppingtrueDetener si no mejora

Monitorear progreso

bash

curl -X GET https://api.astrapp.lat/api/v1/projects/$PROJ_ID/training/$JOB_ID \
  -H "Authorization: Bearer sk_live_xxxxxxxxxxxxx"

json

{
  "job_id": "job_xyz789",
  "status": "training",
  "progress": 45,
  "current_epoch": 5,
  "total_epochs": 10,
  "metrics": {
    "loss": 0.234,
    "accuracy": 0.89
  },
  "estimated_completion": "2024-12-23T11:00:00Z"
}

Estados del entrenamiento

pendingEn cola, esperando recursos

trainingEntrenando activamente

completedExitoso, modelo listo

failedError durante entrenamiento

cancelledCancelado por usuario

Cancelar entrenamiento

bash

curl -X DELETE https://api.astrapp.lat/api/v1/projects/$PROJ_ID/training/$JOB_ID \
  -H "Authorization: Bearer sk_live_xxxxxxxxxxxxx"

Ver logs en tiempo real

Usa Server-Sent Events para ver logs en tiempo real.

bash

curl -X GET https://api.astrapp.lat/api/v1/projects/$PROJ_ID/training/$JOB_ID/logs \
  -H "Authorization: Bearer sk_live_xxxxxxxxxxxxx" \
  -H "Accept: text/event-stream"

Datasets

Inferencia