9 papers
GIFT: Geometry-Informed Low-precision Gradient Communication for LLM Pretraining
Jieying Wang, Shuyuan Fan, Mingkai Zheng +1
Gradient communication is a primary scaling bottleneck in large language model (LLM) pretraining. Communicating gradients in low-precision formats, such as FP8 and NVFP4, can signi…
PHOENIX: Resilient LLM Training with Hot-Swapping via Zero-Overhead Checkpoint
Haotian Xie, Junlin Chen, Mingkai Zheng +2
State-of-the-art large language model (LLM) training takes tens of thousands of graphics processing units (GPUs) for months and encounters failures across the software and hardware…
SCAPE: Accurate and Efficient LLM Training with Extreme Sparse Communication
Mingkai Zheng, Junlin Chen, Haotian Xie +1
Communication increasingly dominates the cost of Large Language Model (LLM) pre-training, especially under data-parallel and sharded training schemes, where gradient synchronizatio…
Reasoning over Precedents Alongside Statutes: Case-Augmented Deliberative Alignment for LLM Safety
Can Jin, Rui Wu, Tong Che +10
Ensuring that Large Language Models (LLMs) adhere to safety principles without refusing benign requests remains a significant challenge. While OpenAI introduces deliberative alignm…
Maple: A Multi-agent System for Portable Deep Learning across Clusters
Molang Wu, Zhao Zhang
Training deep learning (DL) models across Graphics Processing Unit (GPU) clusters is technically challenging. One aspect is that users have to compose command lines to adapt to the…
Pier: Efficient Large Language Model pretraining with Relaxed Global Communication
Shuyuan Fan, Zhao Zhang
Global communication, such as all-reduce and allgather, is the prominent performance bottleneck in large language model (LLM) pretraining. To address this issue, we present Pier, a…