Fine Tuning

Operator uploads a JSONL SFT file where 8% of lines are malformed (missing assistant turn).

Validate the training file format (one chat per line with the required roles) before creating the job; the API surfaces validation errors but pre-checking saves a failed job. Fix or drop malformed lines.

Operator sets n_epochs=20 on a small dataset and the model memorizes/overfits.

Start from auto/default hyperparameters and tune n_epochs, learning_rate_multiplier, and batch_size based on validation loss; high epochs on small data overfit.

Operator fine-tunes with no validation file and reports training loss as quality.

Provide a held-out validation_file to track validation loss; training loss alone overfits. Evaluate the resulting model on a separate eval set before production.