The basic fine-tuning experience is good,
documentation could be expanded to cover more edge cases and advanced custom dataset formatting (keeping documentation sync in such a rapidly changing environment is, of course, a separate issue). Additionally, providing smoother, out-of-the-box integration guides or scripts for directly importing and deploying fine-tuned models to service frameworks like vLLM or Ollama could make the end-to-end workflow much more complete.
Hugging Face
Ollama