4 papers
ES-dLLM: Efficient Inference for Diffusion Large Language Models by Early-Skipping
Zijian Zhu, Fei Ren, Zhanhong Tan +1
Diffusion large language models (dLLMs) are emerging as a promising alternative to autoregressive models (ARMs) due to their ability to capture bidirectional context and the potent…
Concertina: Data-Centric Adaptive Pipeline Parallelism for Efficient Heterogeneous Long-Context LLM Training
Shiju Wang, Yujie Wang, Ao Sun +6
Long context training is crucial for extending LLM context windows. Existing schemes, such as sequence parallelism, incur substantial communication overhead. Pipeline parallelism (…
SoMa: Identifying, Exploring, and Understanding the DRAM Communication Scheduling Space for DNN Accelerators
Jingwei Cai, Xuan Wang, Mingyu Gao +5
Modern Deep Neural Network (DNN) accelerators are equipped with increasingly larger on-chip buffers to provide more opportunities to alleviate the increasingly severe DRAM bandwidt…
Cocco: Hardware-Mapping Co-Exploration towards Memory Capacity-Communication Optimization
Zhanhong Tan, Zijian Zhu, Kaisheng Ma
Memory is a critical design consideration in current data-intensive DNN accelerators, as it profoundly determines energy consumption, bandwidth requirements, and area costs. As DNN…