kandpal-2023-model-size-4x-accuracy-105-docs

IN premise — summaries/2026/08/24/kandpal-2023-long-tail-knowledge-s3-lm-accuracy-depends-on-relevant.md

Created 2026-08-25T02:58:06+00:00

Larger BLOOM models show approximately 4× higher QA accuracy than smaller BLOOM models on questions with more than 10^5 relevant documents.

Summary

When a question requires pulling from a very large body of documents (over 100,000), picking a larger BLOOM model doesn't just give a marginal improvement — it roughly quadruples the accuracy compared to a smaller model. In practical terms, this means that for high-volume, complex question-answering workloads, investing in model scale is one of the highest-leverage choices you can make, since no amount of prompt engineering or retrieval tuning on a small model will close that gap.