Ashwinee Panda

DynaGuard: A Dynamic Guardian Model With User-Defined Policies

Identifying and Evaluating Inactive Heads in Pretrained LLMs

Low Rank Experts Enable Specialization In Dense Transformers

Safety Alignment Should be Made More Than Just a Few Tokens Deep

Dense Backpropagation Improves Training for Sparse Mixture-of-Experts

Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models

Privacy Auditing of Large Language Models

Gemstones: A Model Suite for Multi-Faceted Scaling Laws

LoRI: Reducing Cross-Task Interference in Multi-Task Low-Rank Adaptation

Dense Backpropagation Improves Routing for Sparsely-Gated Mixture-of-Experts

Lottery Ticket Adaptation: Mitigating Destructive Interference in LLMs

Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models

StructMoE: Augmenting MoEs with Hierarchically Routed Low Rank Experts