Search papers, labs, and topics across Lattice.
This paper introduces LLMSurgeon, a framework for estimating the domain-level distribution of an LLM's pretraining corpus by analyzing its generated text. LLMSurgeon formulates this as an inverse problem, estimating a calibrated soft confusion matrix to correct systematic domain confusion and recover the latent mixture prior. The method is evaluated on LLMScan, a new evaluation suite built from open-source LLMs with known pretraining mixtures, demonstrating high fidelity in recovering domain mixtures.
Uncover a model's "digital DNA" – its pretraining data mixture – from its outputs alone, even without access to the training data.
The pretraining data mixture of Large Language Models (LLMs) constitutes their"digital DNA", shaping model behaviors, capabilities, and failure modes. Yet this composition is rarely disclosed, making post-hoc auditing of data combination or provenance difficult. In this work, we formalize $\textbf{{Data Mixture Surgery (DMS)}}$: given only generated text from a target LLM, estimate the domain-level distribution of its pretraining corpus under a predefined taxonomy. We propose $\textbf{{LLMSurgeon}}$, a strong framework that casts DMS as an inverse problem under the label-shift assumption. Rather than directly aggregating classifier outputs, LLMSurgeon estimates a calibrated $\textit{soft}$ confusion matrix and solves a constrained inverse problem to correct systematic domain confusion and recover the latent mixture prior. To evaluate, we introduce $\textbf{{LLMScan}}$, a recipe-verifiable evaluation suite built from open-source LLMs with transparent pretraining mixtures. Across LLMScan, LLMSurgeon recovers domain mixtures with high fidelity under fixed protocols. Our work presents a practical, post-hoc approach for auditing the digital DNA of foundation models without access to their training data.
