Guanhua Wang

Guanhua Wang, Heyang Qin, Sam Ade Jacobs et al.

Zero Redundancy Optimizer (ZeRO) has been used to train a wide range of large language models on massive GPUs clusters due to its ease of use, efficiency, and good scalability. However, when training on low-bandwidth clusters, or at scale which forces batch size per GPU to be small, ZeRO's effective throughput is limited because of high communication volume from gathering weights in forward pass, backward pass, and averaging gradients. This paper introduces three communication volume reduction techniques, which we collectively refer to as ZeRO++, targeting each of the communication collectives in ZeRO. First is block-quantization based all-gather. Second is data remapping that trades-off communication for more memory. Third is a novel all-to-all based quantized gradient averaging paradigm as replacement of reduce-scatter collective, which preserves accuracy despite communicating low precision data. Collectively, ZeRO++ reduces communication volume of ZeRO by 4x, enabling up to 2.16x better throughput at 384 GPU scale.

Marah Abdin, Jyoti Aneja, Hany Awadalla et al. · microsoft-research, stanford

We introduce phi-3-mini, a 3.8 billion parameter language model trained on 3.3 trillion tokens, whose overall performance, as measured by both academic benchmarks and internal testing, rivals that of models such as Mixtral 8x7B and GPT-3.5 (e.g., phi-3-mini achieves 69% on MMLU and 8.38 on MT-bench), despite being small enough to be deployed on a phone. Our training dataset is a scaled-up version of the one used for phi-2, composed of heavily filtered publicly available web data and synthetic data. The model is also further aligned for robustness, safety, and chat format. We also provide parameter-scaling results with a 7B, 14B models trained for 4.8T tokens, called phi-3-small, phi-3-medium, both significantly more capable than phi-3-mini (e.g., respectively 75%, 78% on MMLU, and 8.7, 8.9 on MT-bench). To enhance multilingual, multimodal, and long-context capabilities, we introduce three models in the phi-3.5 series: phi-3.5-mini, phi-3.5-MoE, and phi-3.5-Vision. The phi-3.5-MoE, a 16 x 3.8B MoE model with 6.6 billion active parameters, achieves superior performance in language reasoning, math, and code tasks compared to other open-source models of similar scale, such as Llama 3.1 and the Mixtral series, and on par with Gemini-1.5-Flash and GPT-4o-mini. Meanwhile, phi-3.5-Vision, a 4.2 billion parameter model derived from phi-3.5-mini, excels in reasoning tasks and is adept at handling both single-image and text prompts, as well as multi-image and text prompts.

Guanhua Wang, Chengming Zhang, Zheyu Shen et al.

Given the popularity of generative AI, Large Language Models (LLMs) often consume hundreds or thousands of GPUs for parallelizing and accelerating the training process. Communication overhead becomes more pronounced when training LLMs at scale. To eliminate communication overhead in distributed LLM training, we propose Domino, which provides a generic scheme to hide communication behind computation. By breaking data dependency of a single batch training into smaller independent pieces, Domino pipelines these independent pieces training and provides generic strategy of fine-grained communication and computation overlapping. Extensive results show that, comparing with Megatron-LM, Domino achieves up to 1.3x speedup for LLM training on Nvidia DGX-H100 GPUs.

Guanhua Wang, Olatunji Ruwase, Bing Xie et al.

Model checkpoints are critical Deep Learning (DL) artifacts that enable fault tolerance for training and downstream applications, such as inference. However, writing checkpoints to persistent storage, and other I/O aspects of DL training, are mostly ignored by compute-focused optimization efforts for faster training of rapidly growing models and datasets. Towards addressing this imbalance, we propose FastPersist to accelerate checkpoint creation in DL training. FastPersist combines three novel techniques: (i) NVMe optimizations for faster checkpoint writes to SSDs, (ii) efficient write parallelism using the available SSDs in training environments, and (iii) overlapping checkpointing with independent training computations. Our evaluation using real world dense and sparse DL models shows that FastPersist creates checkpoints in persistent storage up to 116x faster than baseline, and enables per-iteration checkpointing with negligible overhead.

Hongjia Yang, Guanhua Wang, Ziyu Li et al.

Magnetic resonance imaging (MRI) has significantly benefited from the resurgence of artificial intelligence (AI). By leveraging AI's capabilities in large-scale optimization and pattern recognition, innovative methods are transforming the MRI acquisition workflow, including planning, sequence design, and correction of acquisition artifacts. These emerging algorithms demonstrate substantial potential in enhancing the efficiency and throughput of acquisition steps. This review discusses several pivotal AI-based methods in neuro MRI acquisition, focusing on their technological advances, impact on clinical practice, and potential risks.

Ashkan Yousefpour, Brian Q. Nguyen, Siddartha Devic et al.

Federated Learning aims to train distributed deep models without sharing the raw data with the centralized server. Similarly, in distributed inference of neural networks, by partitioning the network and distributing it across several physical nodes, activations and gradients are exchanged between physical nodes, rather than raw data. Nevertheless, when a neural network is partitioned and distributed among physical nodes, failure of physical nodes causes the failure of the neural units that are placed on those nodes, which results in a significant performance drop. Current approaches focus on resiliency of training in distributed neural networks. However, resiliency of inference in distributed neural networks is less explored. We introduce ResiliNet, a scheme for making inference in distributed neural networks resilient to physical node failures. ResiliNet combines two concepts to provide resiliency: skip hyperconnection, a concept for skipping nodes in distributed neural networks similar to skip connection in resnets, and a novel technique called failout, which is introduced in this paper. Failout simulates physical node failure conditions during training using dropout, and is specifically designed to improve the resiliency of distributed neural networks. The results of the experiments and ablation studies using three datasets confirm the ability of ResiliNet to provide inference resiliency for distributed neural networks.

Guanhua Wang, Shivaram Venkataraman, Amar Phanishayee et al.

Model parameter synchronization across GPUs introduces high overheads for data-parallel training at scale. Existing parameter synchronization protocols cannot effectively leverage available network resources in the face of ever increasing hardware heterogeneity. To address this, we propose Blink, a collective communication library that dynamically generates optimal communication primitives by packing spanning trees. We propose techniques to minimize the number of trees generated and extend Blink to leverage heterogeneous communication channels for faster data transfers. Evaluations show that compared to the state-of-the-art (NCCL), Blink can achieve up to 8x faster model synchronization, and reduce end-to-end training time for image classification tasks by up to 40%.

Guanhua Wang, Sudipta Chattopadhyay, Arnab Kumar Biswas et al.

Spectre attacks disclosed in early 2018 expose data leakage scenarios via cache side channels. Specifically, speculatively executed paths due to branch mis-prediction may bring secret data into the cache which are then exposed via cache side channels even after the speculative execution is squashed. Symbolic execution is a well-known test generation method to cover program paths at the level of the application software. In this paper, we extend symbolic execution with modelingof cache and speculative execution. Our tool KLEESPECTRE, built on top of the KLEE symbolic execution engine, can thus provide a testing engine to check for the data leakage through cache side-channel as shown via Spectre attacks. Our symbolic cache model can verify whether the sensitive data leakage due to speculative execution can be observed by an attacker at a given program point. Our experiments show that KLEESPECTREcan effectively detect data leakage along speculatively executed paths and our cache model can further make the leakage detection much more precise.

Ba Dung Le, Guanhua Wang, Mehwish Nasim et al.

Preventing organizations from Cyber exploits needs timely intelligence about Cyber vulnerabilities and attacks, referred as threats. Cyber threat intelligence can be extracted from various sources including social media platforms where users publish the threat information in real time. Gathering Cyber threat intelligence from social media sites is a time consuming task for security analysts that can delay timely response to emerging Cyber threats. We propose a framework for automatically gathering Cyber threat intelligence from Twitter by using a novelty detection model. Our model learns the features of Cyber threat intelligence from the threat descriptions published in public repositories such as Common Vulnerabilities and Exposures (CVE) and classifies a new unseen tweet as either normal or anomalous to Cyber threat intelligence. We evaluate our framework using a purpose-built data set of tweets from 50 influential Cyber security related accounts over twelve months (in 2018). Our classifier achieves the F1-score of 0.643 for classifying Cyber threat tweets and outperforms several baselines including binary classification models. Our analysis of the classification results suggests that Cyber threat relevant tweets on Twitter do not often include the CVE identifier of the related threats. Hence, it would be valuable to collect these tweets and associate them with the related CVE identifier for cyber security applications.

Guanhua Wang, Sudipta Chattopadhyay, Ivan Gotovchits et al.

The Spectre vulnerability in modern processors has been widely reported. The key insight in this vulnerability is that speculative execution in processors can be misused to access the secrets. Subsequently, even though the speculatively executed instructions are squashed, the secret may linger in micro-architectural states such as cache, and can potentially be accessed by an attacker via side channels. In this paper, we propose oo7, a static analysis approach that can mitigate Spectre attacks by detecting potentially vulnerable code snippets in program binaries and protecting them against the attack by patching them. Our key contribution is to balance the concerns of effectiveness, analysis time and run-time overheads. We employ control flow extraction, taint analysis, and address analysis to detect tainted conditional branches and speculative memory accesses. oo7 can detect all fifteen purpose-built Spectre-vulnerable code patterns, whereas Microsoft compiler with Spectre mitigation option can only detect two of them. We also report the results of a large-scale study on applying oo7 to over 500 program binaries (average binary size 261 KB) from different real-world projects. We protect programs against Spectre attack by selectively inserting fences only at vulnerable conditional branches to prevent speculative execution. Our approach is experimentally observed to incur around 5.9% performance overheads on SPECint benchmarks.