it's a detailed article with all useful information. i fine-tuned gpt3.5 with train and valid sets, i got the result file but i don't know how to analyse it! as the train accuracy column values fluctuate with 0.333 - 0.66 - 0,66-0- 0,66- 1e-05- 0-0.66-0.7-1 . does this mean i need more samples for training ?