Nathalie Baracaldo

GneissWeb: Preparing High Quality Data for LLMs at Scale

EPiC: Towards Lossless Speedup for Reasoning Training through Edge-Preserving CoT Condensation

An Efficient and Modular Framework for Targeted Harm Mitigation in LLMs

MAP: Multi-Human-Value Alignment Palette

The Unlearning Mirage: A Dynamic Framework for Evaluating LLM Unlearning

Mitigating Forgetting in LLM Supervised Fine-Tuning and Preference Learning

Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning