David Ondrej demonstrates using Unsloth Studio to choose a model, a finance dataset and training parameters for a small local fine-tuning runFine-tuning adapts a pretrained model by updating its parameters on examples chosen for a particular task or subject.. He watches training loss and explains that the short demonstration is not a full evaluation of model quality.
He then builds question-and-answer examples from a PDF using a dataset recipe. The example uses a remote model API for generation, while he notes that a local model endpoint is another option.
Ondrej discusses model size and hardware constraints, including an observed failure affecting some larger runs on Apple hardware. The video does not establish that the resulting model is reliable for financial decisions.
Watch on YouTube




