This is one example from the data it’s a classification task with 29 categories which are all numbered labels
{“prompt”:“Web Description: Surgical Scissors 5 5 in Mayo Straight Each\nShort Description: SCISSOR MAYO STRAIGHT BLU\nLong Description: Scissor Mayo Straight Blunt\n\n###\n\n”,“completion”:“16”}
Fine-tune command used:
openai api fine_tunes.create -t {training_file_id} -v {validation_file_id} --compute_classification_metrics --classification_n_classes {unique_completion_count} -m ada
The total records for training is around 3lakhs(training+ validation file) and the model used is ada. I am using such a large file for the first time as before I was training on smaller files. So is there a chance that the number of records is causing problems?
The point to note is the same file was used to fine-tune before but without the compute_classification_metrics and classification_n_classes parameters and it had worked