A study of checkpointing in large scale training of deep neural networks
Guardado en:
| Autores: | , , , , |
|---|---|
| Formato: | artículo preliminar |
| Fecha de Publicación: | 2021 |
| Descripción: | Deep learning (DL) applications are increasingly being deployed on HPC systems to leverage the massive parallelism and computing power of those systems. While significant effort has been put to facilitate distributed training by DL frameworks, fault tolerance has been largely ignored. Checkpoint-restart is a common fault tolerance technique in HPC workloads. In this work, we examine the checkpointing implementation of popular DL platforms. We perform experiments with three state-of-theart DL frameworks common in HPC (Chainer, PyTorch, and TensorFlow). We evaluate the computational cost of checkpointing, file formats and file sizes, the impact of scale, and deterministic checkpointing. Our evaluation shows some critical differences in checkpoint mechanisms and exposes several bottlenecks in existing checkpointing implementations. We provide discussion points that can aid users in selecting a fault-tolerant framework to use in HPC. We also provide take-away points that framework developers can use to facilitate better checkpointing of DL workloads in HPC. |
| País: | Repositorio UNA |
| Institución: | Universidad Nacional de Costa Rica |
| Repositorio: | Repositorio UNA |
| Lenguaje: | Inglés |
| OAI Identifier: | oai:null:11056/26772 |
| Acceso en línea: | http://hdl.handle.net/11056/26772 https://doi.org/10.48550/arXiv.2012.00825 |
| Palabra clave: | APRENDIZAJE PROFUNDO RESILIENCIA REDES NEURONALES COMPUTACIÓN DE ALTO RENDIMIENTO DEEP LEARNING RESILIENCE NEURAL NETWORKS HIGH PERFORMANCE COMPUTING |