huggingface.co has a lot of pre-trained models.
any dataset with less that 50k entries should take mostly a day to train.
Ollama has an API interface for LLMs.
Prompt tuning: Is the process of fine tuning a LLM with Q&A style prompts/texts.
RAG (Retrieval-Augmented Generation): Before a LLM answers, it double-checks with a database to see if the answer it generated is in par with the database.