From dataset to trained model
Once a dataset is ready, the workflow generally follows the same shape regardless of the specific tool or provider. You choose a base model to start from, usually one already reasonably capable at the general task, then set hyperparameters that control the training run, most importantly the learning rate, which controls how large each update to the weights is, and the number of epochs, which is how many times the training process passes over the full dataset.
These settings matter more than they might seem. A learning rate that's too high can cause training to overshoot and destabilize the model's existing capabilities, while one that's too low can mean the model barely learns the new pattern at all within a reasonable number of epochs. Most fine-tuning tools ship with sensible defaults for these values precisely because getting them right from scratch is genuinely difficult, and starting from a proven default and adjusting cautiously is usually wiser than guessing.