token budget (you train until run out of money! )
DAMPTP LLM course material see
bluedotimpact practical
Small language models (SLMs) for low resource languages
SLMs finetuned on specific use cases for finance/healthcare
SLMs/guardrails > train on curated datasets so that they remain on task remain on task do not hallucinate cannot be jailbroken. Use SLMs for these tasks (trained for a specific task and on curated datasets)
Training AI models on private data on-premise (healthcare, NASA, space, government sectors, defense, TREs [trusted research environments] etc.)
[LLMs from scratch Vizuara]
LLM from scratch and here
Test early with real people and iterate.