Ravid Shwartz-Ziv's research on model compression, task-aware quantization, efficient representations, and reducing AI memory and computation.
Reducing verbatim loops during generation by penalizing tokens that continue a previously seen sequence.
Encouraging varied text by occasionally excluding dominant choices when several plausible next tokens are available.
Making pretrained transformer attention sensitive to uncertainty during inference.
Compressing an LLM for the task it actually needs to perform, allocating precision to the layers that matter.
Reducing parameter costs with linear transformations that preserve the structure of multidimensional inputs.