Skip to yearly menu bar Skip to main content


Session

Parallel and Distributed Systems 1: Parallelism

Ballroom C

Moderator: Zhihao Jia

Abstract:
Chat is not available.


Ballroom B - Position 1
PipeFisher: Efficient Training of Large Language Models Using Pipelining and Fisher Information Matrices

Kazuki Osawa · Shigang Li · Torsten Hoefler

Pipeline parallelism enables efficient training of Large Language Models (LLMs) on large-scale distributed accelerator clusters. Yet, pipeline bubbles during startup and tear-down reduce the utilization of accelerators. Although efficient pipeline schemes with micro-batching and bidirectional pipelines have been proposed to maximize utilization, a significant number of bubbles cannot be filled using synchronous forward and backward passes. To address this problem, we suggest that extra work be assigned to the bubbles to gain auxiliary benefits in LLM training. As an example in this direction, we propose PipeFisher, which assigns the work of K-FAC, a second-order optimization method based on the Fisher information matrix, to the bubbles to accelerate convergence. In Phase 1 pretraining of BERT-Base and -Large models, PipeFisher reduces the (simulated) training time to 50-75% compared to training with a first-order optimizer by greatly improving the accelerator utilization and benefiting from the improved convergence by K-FAC.


Ballroom B - Position 2
Tutel: Adaptive Mixture-of-Experts at Scale

Changho Hwang · Wei Cui · Yifan Xiong · Ziyue Yang · Ze Liu · Han Hu · Zilong Wang · Rafael Salas · Jithin Jose · Prabhat Ram · HoYuen Chau · Peng Cheng · Fan Yang · Mao Yang · Yongqiang Xiong

Sparsely-gated mixture-of-experts (MoE) has been widely adopted to scale deep learning models to trillion-plus parameters with fixed computational cost. The algorithmic performance of MoE relies on its token routing mechanism that forwards each input token to the right sub-models or experts. While token routing dynamically determines the amount of expert workload at runtime, existing systems suffer inefficient computation due to their static execution, namely static parallelism and pipelining, which does not adapt to the dynamic workload.We present Tutel, a highly scalable stack design and implementation for MoE with dynamically adaptive parallelism and pipelining. Tutel designs an identical layout for distributing MoE model parameters and input data, which can be leveraged by switchable parallelism and dynamic pipelining methods without mathematical inequivalence or tensor migration overhead. This enables adaptive parallelism/pipelining optimization at zero cost during runtime. Based on this key design, Tutel also implements various MoE acceleration techniques including Flexible All-to-All, two-dimensional hierarchical (2DH) All-to-All, fast encode/decode, etc. Aggregating all techniques, Tutel finally delivers 4.96x and 5.75x speedup of a single MoE layer over 16 and 2,048 A100 GPUs, respectively, over the previous state-of-the-art.Our evaluation shows that Tutel efficiently and effectively runs a real-world MoE-based model named SwinV2-MoE, built upon Swin Transformer V2, a state-of-the-art computer vision architecture. On efficiency, Tutel accelerates SwinV2-MoE, achieving up to 1.55x and 2.11x speedup in training and inference over Fairseq, respectively. On effectiveness, the SwinV2-MoE model achieves superior accuracy in both pre-training and down-stream computer vision tasks such as COCO object detection than the counterpart dense model, indicating the readiness of Tutel for end-to-end real-world model training and inference.


Ballroom B - Position 3
Breadth-First Pipeline Parallelism

Joel Lamy-Poirier

We introduce Breadth-First Pipeline Parallelism, a novel training schedule which optimizes the combination of pipeline and data parallelism. Breadth-First Pipeline Parallelism lowers training time, cost and memory usage by combining a high GPU utilization with a small batch size per GPU, and by making use of fully sharded data parallelism. Experimentally, we observed an increase of up to 43% in training throughput for a 52 billion-parameter model using a small batch size per GPU compared to Megatron-LM, which would reduce the training time and cost by the same amount on a large GPU cluster.