Ask HN: Most efficient way to fine-tune an LLM in 2024?
In Apr 2024 what is the most efficient way to fine-tune an LLM?
In particular we are trying to understand performance vs. cost trade-offs. We don't have a budget to train from scratch.
We are working with a proprietary data set on the order of 100M tokens and are looking to fine-tune a general purpose language model and also create task-specific models based on the same corpus.
In particular we are trying to understand performance vs. cost trade-offs. We don't have a budget to train from scratch.
We are working with a proprietary data set on the order of 100M tokens and are looking to fine-tune a general purpose language model and also create task-specific models based on the same corpus.
Any help would be appreciated!