Distributional Adversarial Training utilizes cont. adv. training on diffusion-based adv. examples to close a gap in population-robust risk estimation.
-
ASSELab/DAT-Qwen2.5-14B-Instruct
Text Generation • 15B • Updated • 15 -
ASSELab/Diffusion-Llama-3-8B-Instruct
Text Generation • 8B • Updated • 16 -
ASSELab/DAT-Llama-3-8B-Instruct
Text Generation • 8B • Updated • 35 • 1 -
Closing the Distribution Gap in Adversarial Training for LLMs
Paper • 2602.15238 • Published