AI2's Nathan Lambert argues supervised fine-tuning lacks structured literature as new study tests models up to 235B parameters

1/ We fine-tune a lot of customer models, so we decided to systematically try and figure out some best practices for finetuning. SFT isn't sexy, but it's still important. We vary one SFT lever at a time across 2 model families, dense + MoE to 235B, on 4 real-world customer datasets.

What makes this clean is that each dataset is paired with an eval that took weeks to build with the customer, and the training outputs were generated to pass that eval. So the supervised target and the thing we measure downstream are the same criterion, which strips out the usual confounders