arXiv subjects

Siling Yang

Publications and source records attributed to Siling Yang.

2 recordsLinked to original sources

arXiv2024

Frenzy: A Memory-Aware Serverless LLM Training System for Heterogeneous GPU Clusters

Existing work only effective on a given number of GPUs, often neglecting the complexities involved in manually determining the specific types and quantities of GPUs needed, which can be a significant burden for developers. To address this issue, we propose Frenzy, a memory-aware serverless computing method for heterogeneous GPU clusters. Frenzy allows users to submit models without worrying about underlying hardware resources. First, Frenzy predicts the required number and type of GPUs by estimating the GPU memory usage of the LLM. Then, it employs a low-overhead heterogeneity-aware scheduling method to optimize training efficiency. We validated Frenzy's performance by conducting multi-task LLM training tests on a heterogeneous GPU cluster with three different GPU types. The results show that Frenzy's memory usage prediction accuracy exceeds 92\%, the scheduling overhead is reduced by 10 times, and it reduces the average job completion time by 12\% to 18\% compared to state-of-the-art methods.

cs.DC↗

arXiv2024

Optimizing Large Model Training through Overlapped Activation Recomputation

Large model training often uses recomputation to alleviate memory pressure and pipelines to exploit the parallelism of data, tensors, and devices. However, existing recomputation approaches may incur high overhead when training real-world models, as they are executed on demand in the critical training path. In this paper, we present Lynx, a new recomputation framework to reduce overhead by overlapping recomputation with communication in training pipelines. To reduce the large search space for recomputation strategies, we propose a heuristic-based recomputation scheduling algorithm, which is based on the observation that there are identical structures in large DNN models so that we can apply the same scheduling policy to all such structures. Additionally, we propose a recomputation-aware model partitioning method to balance each stage's execution time for improved training throughput. Our comprehensive evaluation using GPT models with 1.3B-23B parameters shows that Lynx outperforms existing recomputation approaches by up to 1.37x.

cs.DC↗