analiza aiqbench
The GRADIEND package offers a practical, open-source implementation of gradient-based feature learning methods within large language models, streamlining workflows from data creation to persistent model rewriting. By enabling direct manipulation of factual-counterfactual gradients, this tool facilitates deeper interpretability and controlled editing of semantic representations without relying on black-box inference. This development represents a significant step forward in making advanced mechanistic interpretability techniques accessible for researchers working on model transparency and feature alignment.
The GRADIEND Python Package: An End-to-End System for Gradient-Based Feature Learning
arXiv:2602.23993v3 Announce Type: replace Abstract: We present gradiend, an open-source Python package that operationalizes the GRADIEND method for learning feature directions from factual-counterfactual MLM and CLM gradients in language models. The package provides a unified workflow for feature-related data creation, training, evaluation, visualization, persistent model rewriting via controlled weight updates, and multi-feature comparison. We demonstrate gradiend through an English pronoun running example, a semantic sentiment use case that evaluates lexical generalization to held-out target words, and a large-scale feature comparison.
Abstract:We present gradiend, an open-source Python package that operationalizes the GRADIEND method for learning feature directions from factual-counterfactual MLM and CLM gradients in language models. The package provides a unified workflow for feature-related data creation, training, evaluation, visualization, persistent model rewriting via controlled weight updates, and multi-feature comparison. We demonstrate gradiend through an English pronoun running example, a semantic sentiment use case that evaluates lexical generalization to held-out target words, and a large-scale feature comparison.
Computation and Language (cs.CL)
arXiv:2602.23993 [cs.CL]
(or arXiv:2602.23993v3 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2602.23993 Focus to learn more arXiv-issued DOI via DataCite
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 · 168 cuvinte
