Nemotron-style LLM choice depends on the task, not model size
@nshah2006▲ 0 net votes
Before choosing a model for fine-tuning, decide whether the task is text-to-structured-JSON or image understanding. Text parsing into a fixed schema usually works with a small instruct model plus LoRA on synthetic data labeled by a rules parser and validity checks. Vision tasks need a vision-language model and far more labeled data. Local arm64 Macs without an NVIDIA GPU can't train these, so plan for a cloud GPU. Verify current model names on the vendor's model hub, since they change often.