analiza aiqbench
Influcoder addresses the computational bottleneck of influence functions in large language models by distilling gradient rankings into an encoder, enabling scalable and efficient data attribution. This advancement is critical for practical dataset curation tasks such as identifying toxic training samples without prohibitive storage or processing costs. The method represents a significant step toward making rigorous model auditing feasible at scale.
Influcoder: Distilling Decoders' Gradient Influence Rankings into an Encoder for Data Attribution
arXiv:2606.13668v1 Announce Type: new Abstract: With the growth of LLMs' (Large Language Models) capabilities, there has been an increasing push to curate high quality datasets by filtering samples in the training data. In general, Data Attribution (DA) methods aim to estimate how individual samples in a training dataset can precondition a model to generate certain outputs. As an example, one might be interested in which samples in the data could be the source of toxic behavior after training the LLM. Many methods quantify this conditioning through the paradigm of influence functions. While methods of this family are effective in its function, they lack the necessary processing speed and storage compactness to be practically implemented on large datasets. We propose a method, Influcoder, as a quick and cost-effective approach to influence-based Data Attribution at scale.
Abstract:With the growth of LLMs' (Large Language Models) capabilities, there has been an increasing push to curate high quality datasets by filtering samples in the training data. In general, Data Attribution (DA) methods aim to estimate how individual samples in a training dataset can precondition a model to generate certain outputs. As an example, one might be interested in which samples in the data could be the source of toxic behavior after training the LLM. Many methods quantify this conditioning through the paradigm of influence functions. While methods of this family are effective in its function, they lack the necessary processing speed and storage compactness to be practically implemented on large datasets. We propose a method, Influcoder, as a quick and cost-effective approach to influence-based Data Attribution at scale.
Computation and Language (cs.CL)
arXiv:2606.13668 [cs.CL]
(or arXiv:2606.13668v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.13668 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 224 cuvinte