Study Traces Where Fine-Tuned Behaviors Reside in Language Models via KV-Cache Patching
A new arXiv paper introduces KV-Cache Patching, a method to trace how and where fine-tuned behaviors are instantiated in language models, evaluating which parts of the model change after fine-tuning.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-02 05:00 · arXiv cs.CL
KV-Cache Patching: Tracing How and Where Fine-Tuned Behaviors Arise