TrainingMetricsSchema

end_train_messagestring[]

Messages describing how the training job concluded, such as "Training job ended" or an error description.

end_train_statusstring[]

Final status of the training job: dormant (completed and resources released) or failed.

eval_lossnumber[]

Validation loss values, typically the loss before fine-tuning followed by the loss after fine-tuning. Lower values indicate better generalization.

eval_perplexitynumber[]

Validation perplexity values before and after fine-tuning. May be empty if perplexity is not computed.

lossnumber[]

Training loss values recorded across training steps; the final value is the final training loss.

perplexitynumber[]

Training perplexity values recorded across steps. May be empty if perplexity is not computed.

step_loss_validation[]

Validation loss values recorded at each evaluation step during training.

TrainingMetricsSchema

{
  "end_train_message": [
    "string"
  ],
  "end_train_status": [
    "string"
  ],
  "eval_loss": [
    0
  ],
  "eval_perplexity": [
    0
  ],
  "loss": [
    0
  ],
  "perplexity": [
    0
  ],
  "step_loss_validation": [
    null
  ]
}