NanoKnow: How to Know What Your Language Model Knows

TL;DR

NanoKnow introduces a benchmark dataset to analyze how large language models rely on pre-training data versus external evidence for answering questions, revealing insights into knowledge encoding and model behavior.

Contribution

It provides a transparent dataset and methodology to disentangle pre-training and external knowledge sources in LLMs, advancing understanding of their knowledge mechanisms.

Findings

Answer frequency in pre-training data strongly influences accuracy.

External evidence can reduce reliance on pre-training data.

Models perform better on answers seen during pre-training, even with external evidence.

Abstract

How do large language models (LLMs) know what they know? Answering this question has been difficult because pre-training data is often a "black box" - unknown or inaccessible. The recent release of nanochat - a family of small LLMs with fully open pre-training data - addresses this as it provides a transparent view into where a model's parametric knowledge comes from. Towards the goal of understanding how knowledge is encoded by LLMs, we release NanoKnow, a benchmark dataset that partitions questions from Natural Questions and SQuAD into splits based on whether their answers are present in nanochat's pre-training corpus. Using these splits, we can now properly disentangle the sources of knowledge that LLMs rely on when producing an output. To demonstrate NanoKnow's utility, we conduct experiments using eight nanochat checkpoints. Our findings show: (1) closed-book accuracy is strongly influenced by answer frequency in the pre-training data, (2) providing external evidence can mitigate this frequency dependence, (3) even with external evidence, models are more accurate when answers were seen during pre-training, demonstrating that parametric and external knowledge are complementary, and (4) non-relevant information is harmful, with accuracy decreasing based on both the position and the number of non-relevant contexts. We release all NanoKnow artifacts at https://github.com/castorini/NanoKnow.