4 papers
Parallelism Strategy Chaining for Fast Training Convergence
Minchul Kang, Changyong Shin, Younghun Go +4
Selecting a parallelism strategy - the configuration of data, tensor, and pipeline parallelism degrees together with micro- and global-batch sizes - largely determines the training…
Enabling KV Caching of Shared Prefix for Diffusion Language Models
Younghun Go, Jaehoon Han, Changyong Shin +2
Key-value (KV) caching for shared prefixes is essential for high-throughput large language model (LLM) serving, but it faces critical challenges in emerging diffusion language mode…
Training Time Prediction for Mixed Precision-based Distributed Training
Minchul Kang, Changyong Shin, Jinwoo Jeong +5
Accurate prediction of training time in distributed deep learning is crucial for resource allocation, cost estimation, and job scheduling. We observe that the floating-point precis…
GPU Memory Prediction for Multimodal Model Training
Jinwoo Jeong, Minchul Kang, Younghun Go +5
As deep learning models in agentic AI systems grow in scale and complexity, GPU memory requirements increase and often exceed the available GPU memory capacity, so that out-of-memo…