The success of Transformer models has pushed the deep learning model scale to billions of parameters, but the memory limitation of a single GPU has led to an urgent need for training on multi-GPU clusters. However, the best practice for choosing the optimal parallel strategy is still lacking, as it requires domain expertise in both deep learning and parallel computing. The Colossal-AI system addressed the above challenge by introducing a unified interface to scale your sequential code of model training to distributed environments. It supports parallel training methods such as data, pipeline, tensor, and sequence parallelism and is integrated with heterogeneous training and zero redundancy optimizer. Compared to the baseline system, Colossal-AI can achieve up to 2.76 times training speedup on large-scale models.

Shenggui Li, Fuzhao Xue, Chaitanya Baranwal, Yongbin Li, Yang You. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Chenhui Shen, Liying Cheng, Xuan-Phi Nguyen, Yang You, Lidong Bing. Findings of the Association for Computational Linguistics: EMNLP 2023. 2023.

We propose a deep learning approach for directly estimating relative atmospheric visibility from outdoor photos without relying on weather images or data that require expensive sensing or custom capture. Our data-driven approach capitalizes on a large collection of Internet images to learn rich scene and visibility varieties. The relative CNN-RNN coarse-to-fine model, where CNN stands for convolutional neural network and RNN stands for recurrent neural network, exploits the joint power of relative support vector machine, which has a good ranking representation, and the data-driven deep learning features derived from our novel CNN-RNN model. The CNN-RNN model makes use of shortcut connections to bridge a CNN module and an RNN coarse-to-fine module. The CNN captures the global view while the RNN simulates human's attention shift, namely, from the whole image (global) to the farthest discerned region (local). The learned relative model can be adapted to predict absolute visibility in limited scenarios. Extensive experiments and comparisons are performed to verify our method. We have built an annotated dataset consisting of about 40000 images with 0.2 million human annotations. The large-scale, annotated visibility data set will be made available to accompany this paper.

Since its creation, the ImageNet-1k benchmark set has played a significant role as a benchmark for ascertaining the accuracy of different deep neural net (DNN) models on the image classification problem. Moreover, in recent years it has also served as the principal benchmark for assessing different approaches to DNN training. Finishing a 90-epoch ImageNet-1k training with ResNet-50 on a NVIDIA M40 GPU takes 14 days. This training requires $10^{18}$ single precision operations in total. On the other hand, the world's current fastest supercomputer can finish $3 \times 10^{17}$ single precision operations per second (according to the Nov 2018 Top 500 results). If we can make full use of the computing capability of the fastest supercomputer, we should be able to finish the training in several seconds. Over the last two years, researchers have focused on closing this significant performance gap through scaling DNN training to larger numbers of processors. Most successful approaches to scaling ImageNet training have used the synchronous mini-batch stochastic gradient descent (SGD). However, to scale synchronous SGD one must also increase the batch size used in each iteration. Thus, for many researchers, the focus on scaling DNN training has translated into a focus on developing training algorithms that enable increasing the batch size in data-parallel synchronous SGD without losing accuracy over a fixed number of epochs. In this paper, we investigate supercomputers' capability of speeding up DNN training. Our approach is to use a large batch size, powered by the Layer-wise Adaptive Rate Scaling (LARS) algorithm, for efficient usage of massive computing resources. Our approach is generic, as we empirically evaluate the effectiveness on five neural networks: AlexNet, AlexNet-BN, GNMT, ResNet-50, and ResNet-50-v2 trained with large datasets while preserving the state-of-the-art test accuracy. Compared to the baseline of a previous study from Goyal et al. (1), our approach shows higher test accuracy on batch sizes that are larger than 16K. When we use the same baseline, our results are better than Goyal et al. for all the batch sizes (Fig. 20). Using 2,048 Intel Xeon Platinum 8160 processors, we reduce the 100-epoch AlexNet training time from hours to 11 minutes. With 2,048 Intel Xeon Phi 7250 Processors, we reduce the 90-epoch ResNet-50 training time from hours to 20 minutes. Our implementation is open source and has been released in the Intel distribution of Caffe, Facebook's PyTorch, and Google's TensorFlow. The difference between this paper and the conference-version of our work (2) includes: (1) we implement our approach on Google's cloud Tensor Processing Unit (TPU) platform, which verifies our previous success on CPUs and GPUs. (2) we scale the batch size of ResNet-50-v2 to 32K and achieve 76.3% accuracy, which is better than the 75.3% accuracy achieved in our conference paper. (3) we apply our approach to Google's Neural Machine Translation (GNMT) application, which helps us to achieves $4\times$ speedup on the cloud TPUs.

Dataset distillation reduces the storage and computational consumption of training a network by generating a small surrogate dataset that encapsulates rich information of the original large-scale one. However, previous distillation methods heavily rely on the sample-wise iterative optimization scheme. As the images-per-class (IPC) setting or image resolution grows larger, the necessary computation will demand overwhelming time and resources. In this work, we intend to incorporate generative diffusion techniques for computing the surrogate dataset. Observing that key factors for constructing an effective surrogate dataset are representativeness and diversity, we design additional minimax criteria in the generative training to enhance these facets for the generated images of diffusion models. We present a theoretical model of the process as hierarchical diffusion control demonstrating the flexibility of the diffusion process to target these criteria without jeopardizing the faithfulness of the sample to the desired distribution. The proposed method achieves state-of-the-art validation performance while demanding much less computational resources. Under the 100-IPC setting on Image Woof, our method requires less than one-twentieth the distillation time of previous methods, yet yields even better performance. Source code and generated data are available in https://github.com/vimar-gu/MinimaxDiffusion.