Training Recipe: Strongest Zero-Shot Model Beats General Retrievers

tomaarsen · x · 2026-08-26

The author shares the full recipe for training a strong zero-shot model: using a pre-supervised checkpoint, 1M domain pairs, in-batch negatives, full document length, and a higher-than-usual learning rate. The finetuned model achieves 84.9% rank-1 accuracy, significantly outperforming the zero-shot baseline. Multi-vector/late-interaction architectures are performing exceptionally well.

Related event: Finetuning ColBERT on a single RTX 3090 beats general-purpose retrievers in medical search(13 posts)→

Original post →

More from Apps

Apps channel →