Can you provide a complete training and validation data set sample.
When I provide an emotional dialogue training sample for fine-tuning, the completed model appears messy in the Completions interface, and the answers are always irrelevant.
This is my request parms:
{
“training_file”:“file-xxxx”,
“model”:“davinci:xxxxx-2023-03-23-13-16-39”,
“suffix”: “chat_xxx_”,
“validation_file”: “file-xxxx”,
“n_epochs”: 4,
“batch_size”: 10,
“learning_rate_multiplier”: 0.02,
“prompt_loss_weight”: 0.02,
“compute_classification_metrics”: false
}