speedups
Improvements in computational efficiency achieved through algorithmic changes or hardware advancements, which can drastically reduce training or inference times for AI models.
- A faster training algorithm for regression trees with linear leaves, and an analysis of its complexity
- AutoJudge: Judge Decoding Without Manual Annotation
- CodeGEMM: A Codebook-Centric Approach to Efficient GEMM in Quantized LLMs
- PARALLELPROMPT: Extracting Parallelism from Large Language Model Queries
- SHAP values via sparse Fourier representation