Retune an existing model using the generated synthetic dataset from the source model | Distil labs

curl -X POST https://https://api.distillabs.ai/models/from \
     -H "Content-Type: application/json" \
     -d '{
  "name": "Llama-3.2-1B-Instruct-Retuned",
  "student-model": "Llama-3.2-1B-Instruct",
  "tuning-parameters": {}
}'

{
  "created_at": "2024-06-10T14:45:00Z",
  "id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
  "name": "Llama-3.2-1B-Instruct-Retuned",
  "teacher_evaluation_ids": [
    "7c9e6679-7425-40de-944b-e07fc1f90ae7"
  ],
  "training": {
    "id": "1c9e6679-7425-40de-944b-e07fc1f90ae7",
    "upload_id": "2d9e6679-7425-40de-944b-e07fc1f90ae7"
  },
  "training_evaluation_results": {
    "student-base": {
      "binary": 0.82,
      "llm-as-a-judge": 0.88,
      "meteor": 0.75,
      "rouge": 0.79
    },
    "student-tuned": {
      "binary": 0.89,
      "llm-as-a-judge": 0.92,
      "meteor": 0.8,
      "rouge": 0.85
    },
    "teacher": {
      "binary": 0.95,
      "llm-as-a-judge": 0.97,
      "meteor": 0.9,
      "rouge": 0.93
    }
  },
  "training_status": "JOB_RUNNING",
  "upload_ids": [
    "4e9e6679-7425-40de-944b-e07fc1f90ae7"
  ],
  "evaluation_results": {
    "binary": 0.87,
    "llm-as-a-judge": 0.91,
    "meteor": 0.78,
    "rouge": 0.82
  },
  "task_details": {
    "student_model": "Llama-3.2-1B-Instruct",
    "task": "text-generation",
    "teacher_model": "Llama-3.2-1B-Instruct-Teacher"
  }
}

Retune an existing model using its synthetic dataset. This creates a new model and runs a new finetune job.

Path parameters

fromstringRequiredformat: "uuid"

The model ID to use as a base for the retuning

Request

This endpoint expects an object.

namestringRequiredformat: "^[a-zA-Z0-9-]+$"

Name of the retuned model to be created

student-modelenumRequired

tuning-parametersobjectRequired

Response

Model created

created_atdatetime

idstringformat: "uuid"

namestringformat: "^[a-zA-Z0-9-]+$"

teacher_evaluation_idslist of strings

trainingobject or null

training_evaluation_resultsobject or null

training_statusenum

upload_idslist of strings

evaluation_resultsobject or nullDeprecated

task_detailsobject or null

Path parameters

Request

Response

Errors