Fine-tune Cost Estimator

Dataset size × epochs × model price → total fine-tune cost.

BASE MODEL

TRAINING EXAMPLES

TOKENS / EXAMPLE

EPOCHS

ESTIMATED FINE-TUNE COST

one-time

Total tokens

Inference multiplier

INFERENCE NOT INCLUDED

Before fine-tuning: can you solve it with prompt + few-shot examples? Often you can, at 1% of the cost. Fine-tune only for behavior/format/tone — for adding knowledge use RAG.