Celeritas: Fast Optimizer for Large Dataflow Graphs
This addresses the problem of slow and inefficient model parallelism for researchers and practitioners training large AI models, representing a strong incremental improvement.
The paper tackles the challenge of efficiently training large neural networks via model parallelism by proposing Celeritas, a fast framework for device placement optimization, which reduces placement policy generation time by 26.4% and improves model running time by 34.2% compared to state-of-the-art methods.
The rapidly enlarging neural network models are becoming increasingly challenging to run on a single device. Hence model parallelism over multiple devices is critical to guarantee the efficiency of training large models. Recent proposals fall short either in long processing time or poor performance. Therefore, we propose Celeritas, a fast framework for optimizing device placement for large models. Celeritas employs a simple but efficient model parallelization strategy in the Standard Evaluation, and generates placement policies through a series of scheduling algorithms. We conduct experiments to deploy and evaluate Celeritas on numerous large models. The results show that Celeritas not only reduces the placement policy generation time by 26.4\% but also improves the model running time by 34.2\% compared to most advanced methods.