arXiv subjects

Jiayi Yao

Publications and source records attributed to Jiayi Yao.

23 records · Page 2Linked to original sources

arXiv2024

CacheGen: KV Cache Compression and Streaming for Fast Large Language Model Serving

As large language models (LLMs) take on complex tasks, their inputs are supplemented with longer contexts that incorporate domain knowledge. Yet using long contexts is challenging, as nothing can be generated until the whole context is processed by the LLM. While the context-processing delay can be reduced by reusing the KV cache of a context across different inputs, fetching the KV cache, which contains large tensors, over the network can cause high extra network delays. CacheGen is a fast context-loading module for LLM systems. First, CacheGen uses a custom tensor encoder, leveraging KV cache's distributional properties to encode a KV cache into more compact bitstream representations with negligible decoding overhead, to save bandwidth usage. Second, CacheGen adapts the compression level of different parts of a KV cache to cope with changes in available bandwidth, in order to maintain low context-loading delay and high generation quality. % When available bandwidth drops, CacheGen may raise the compression level for a part of the context or recompute its KV cache on the fly. We test CacheGen on popular LLMs and datasets. Compared to the recent systems that reuse the KV cache, CacheGen reduces the KV cache size by 3.5-4.3x and the total delay in fetching and processing contexts by 3.2-3.7x with negligible impact on the LLM response quality. Our code is at: https://github.com/UChi-JCL/CacheGen.

cs.NI↗

arXiv2023

Fuzzing Automatic Differentiation in Deep-Learning Libraries

Deep learning (DL) has attracted wide attention and has been widely deployed in recent years. As a result, more and more research efforts have been dedicated to testing DL libraries and frameworks. However, existing work largely overlooked one crucial component of any DL system, automatic differentiation (AD), which is the basis for the recent development of DL. To this end, we propose $\nabla$Fuzz, the first general and practical approach specifically targeting the critical AD component in DL libraries. Our key insight is that each DL library API can be abstracted into a function processing tensors/vectors, which can be differentially tested under various execution scenarios (for computing outputs/gradients with different implementations). We have implemented $\nabla$Fuzz as a fully automated API-level fuzzer targeting AD in DL libraries, which utilizes differential testing on different execution scenarios to test both first-order and high-order gradients, and also includes automated filtering strategies to remove false positives caused by numerical instability. We have performed an extensive study on four of the most popular and actively-maintained DL libraries, PyTorch, TensorFlow, JAX, and OneFlow. The result shows that $\nabla$Fuzz substantially outperforms state-of-the-art fuzzers in terms of both code coverage and bug detection. To date, $\nabla$Fuzz has detected 173 bugs for the studied DL libraries, with 144 already confirmed by developers (117 of which are previously unknown bugs and 107 are related to AD). Remarkably, $\nabla$Fuzz contributed 58.3% (7/12) of all high-priority AD bugs for PyTorch and JAX during a two-month period. None of the confirmed AD bugs were detected by existing fuzzers.

cs.SE↗

arXiv2022

A pruning method based on the dissimilarity of angle among channels and filters

Convolutional Neural Network (CNN) is more and more widely used in various fileds, and its computation and memory-demand are also increasing significantly. In order to make it applicable to limited conditions such as embedded application, network compression comes out. Among them, researchers pay more attention to network pruning. In this paper, we encode the convolution network to obtain the similarity of different encoding nodes, and evaluate the connectivity-power among convolutional kernels on the basis of the similarity. Then impose different level of penalty according to different connectivity-power. Meanwhile, we propose Channel Pruning base on the Dissimilarity of Angle (DACP). Firstly, we train a sparse model by GL penalty, and impose an angle dissimilarity constraint on the channels and filters of convolutional network to obtain a more sparse structure. Eventually, the effectiveness of our method is demonstrated in the section of experiment. On CIFAR-10, we reduce 66.86% FLOPs on VGG-16 with 93.31% accuracy after pruning, where FLOPs represents the number of floating-point operations per second of the model. Moreover, on ResNet-32, we reduce FLOPs by 58.46%, which makes the accuracy after pruning reach 91.76%.

cs.CV↗

arXiv2022

Neural Network Panning: Screening the Optimal Sparse Network Before Training

Pruning on neural networks before training not only compresses the original models, but also accelerates the network training phase, which has substantial application value. The current work focuses on fine-grained pruning, which uses metrics to calculate weight scores for weight screening, and extends from the initial single-order pruning to iterative pruning. Through these works, we argue that network pruning can be summarized as an expressive force transfer process of weights, where the reserved weights will take on the expressive force from the removed ones for the purpose of maintaining the performance of original networks. In order to achieve optimal expressive force scheduling, we propose a pruning scheme before training called Neural Network Panning which guides expressive force transfer through multi-index and multi-process steps, and designs a kind of panning agent based on reinforcement learning to automate processes. Experimental results show that Panning performs better than various available pruning before training methods.

cs.LG↗

arXiv2022

Automated Testing of Image Captioning Systems

Image captioning (IC) systems, which automatically generate a text description of the salient objects in an image (real or synthetic), have seen great progress over the past few years due to the development of deep neural networks. IC plays an indispensable role in human society, for example, labeling massive photos for scientific studies and assisting visually-impaired people in perceiving the world. However, even the top-notch IC systems, such as Microsoft Azure Cognitive Services and IBM Image Caption Generator, may return incorrect results, leading to the omission of important objects, deep misunderstanding, and threats to personal safety. To address this problem, we propose MetaIC, the \textit{first} metamorphic testing approach to validate IC systems. Our core idea is that the object names should exhibit directional changes after object insertion. Specifically, MetaIC (1) extracts objects from existing images to construct an object corpus; (2) inserts an object into an image via novel object resizing and location tuning algorithms; and (3) reports image pairs whose captions do not exhibit differences in an expected way. In our evaluation, we use MetaIC to test one widely-adopted image captioning API and five state-of-the-art (SOTA) image captioning models. Using 1,000 seeds, MetaIC successfully reports 16,825 erroneous issues with high precision (84.9\%-98.4\%). There are three kinds of errors: misclassification, omission, and incorrect quantity. We visualize the errors reported by MetaIC, which shows that flexible overlapping setting facilitates IC testing by increasing and diversifying the reported errors. In addition, MetaIC can be further generalized to detect label errors in the training dataset, which has successfully detected 151 incorrect labels in MS COCO Caption, a standard dataset in image captioning.

cs.SE↗