Search papers, labs, and topics across Lattice.
The paper introduces Weight Patching, a parameter-space intervention method that transfers weights from a behavior-specialized LLM into a base model to isolate the modules responsible for a specific capability. By using a vector-anchor behavioral interface to assess the recovery of task-relevant control states, they identify a hierarchy of modules involved in instruction following, ranging from shallow carriers to downstream execution circuits. The effectiveness of Weight Patching is validated through mechanism-aware model merging, demonstrating improved selective fusion across expert combinations.
Uncover the hidden hierarchy of LLM components driving instruction following, revealing that activation-space importance doesn't always mean parameter-space encoding.
Mechanistic interpretability seeks to localize model behavior to the internal components that causally realize it. Prior work has advanced activation-space localization and causal tracing, but modules that appear important in activation space may merely aggregate or amplify upstream signals rather than encode the target capability in their own parameters. To address this gap, we propose Weight Patching, a parameter-space intervention method for source-oriented analysis in paired same-architecture models that differ in how strongly they express a target capability under the inputs of interest. Given a base model and a behavior-specialized counterpart, Weight Patching replaces selected module weights from the specialized model into the base model under a fixed input. We instantiate the method on instruction following and introduce a framework centered on a vector-anchor behavioral interface that provides a shared internal criterion for whether a task-relevant control state has been formed or recovered in open-ended generation. Under this framework, the analysis reveals a hierarchy from shallow candidate source-side carriers to aggregation and routing modules, and further to downstream execution circuits. The recovered component scores can also guide mechanism-aware model merging, improving selective fusion across the evaluated expert combinations and providing additional external validation.
