MixLoRA: Enhancing Large Language Models Fine-Tuning with LoRA based Mixture of Experts

Large Language Models (LLMs) have showcased exceptional performance across a wide array of Natural Language Processing (NLP) tasks. Fine-tuning techniques are commonly utilized to tailor pre-trained models to specific applications. While methods like LoRA have effectively tackled GPU memory constraints during fine-tuning, their applicability is often restricted to limited performance, especially on multi-task. On the other hand, Mix-of-Expert (MoE) models, such as Mixtral 8x7B, demonstrate remarkable performance across multiple NLP tasks while maintaining a reduced parameter count. However, the resource requirements of these MoEs still challenging, particularly for consumer-grade GPUs only have limited VRAM. To address these challenge, we propose MixLoRA, an innovative approach aimed at constructing a resource-efficient sparse MoE model based on LoRA. MixLoRA inserts multiple LoRA-based experts within the feed-forward network block of a frozen pre-trained dense model through fine-tuning, employing a commonly used top-k router. Unlike other LoRA based MoE methods, MixLoRA enhances model performance by utilizing independently configurable attention-layer LoRA adapters, supporting the use of LoRA and its variants for the construction of experts, and applying auxiliary load balance loss to address the imbalance problem of the router. In experiments, MixLoRA achieves commendable performance across all evaluation metrics in both single-task and multi-task learning scenarios. Implemented within the m-LoRA framework, MixLoRA enables parallel fine-tuning of multiple mixture-of-experts models on a single 24GB consumer-grade GPU without quantization, thereby reducing GPU memory consumption by 41\% and latency during the training process by 17\%.

BadPart: Unified Black-box Adversarial Patch Attacks against Pixel-wise Regression Tasks

Pixel-wise regression tasks (e.g., monocular depth estimation (MDE) and optical flow estimation (OFE)) have been widely involved in our daily life in applications like autonomous driving, augmented reality and video composition. Although certain applications are security-critical or bear societal significance, the adversarial robustness of such models are not sufficiently studied, especially in the black-box scenario. In this work, we introduce the first unified black-box adversarial patch attack framework against pixel-wise regression tasks, aiming to identify the vulnerabilities of these models under query-based black-box attacks. We propose a novel square-based adversarial patch optimization framework and employ probabilistic square sampling and score-based gradient estimation techniques to generate the patch effectively and efficiently, overcoming the scalability problem of previous black-box patch attacks. Our attack prototype, named BadPart, is evaluated on both MDE and OFE tasks, utilizing a total of 7 models. BadPart surpasses 3 baseline methods in terms of both attack performance and efficiency. We also apply BadPart on the Google online service for portrait depth estimation, causing 43.5% relative distance error with 50K queries. State-of-the-art (SOTA) countermeasures cannot defend our attack effectively.

Couler: Unified Machine Learning Workflow Optimization in Cloud

Machine Learning (ML) has become ubiquitous, fueling data-driven applications across various organizations. Contrary to the traditional perception of ML in research, ML workflows can be complex, resource-intensive, and time-consuming. Expanding an ML workflow to encompass a wider range of data infrastructure and data types may lead to larger workloads and increased deployment costs. Currently, numerous workflow engines are available (with over ten being widely recognized). This variety poses a challenge for end-users in terms of mastering different engine APIs. While efforts have primarily focused on optimizing ML Operations (MLOps) for a specific workflow engine, current methods largely overlook workflow optimization across different engines. In this work, we design and implement Couler, a system designed for unified ML workflow optimization in the cloud. Our main insight lies in the ability to generate an ML workflow using natural language (NL) descriptions. We integrate Large Language Models (LLMs) into workflow generation, and provide a unified programming interface for various workflow engines. This approach alleviates the need to understand various workflow engines' APIs. Moreover, Couler enhances workflow computation efficiency by introducing automated caching at multiple stages, enabling large workflow auto-parallelization and automatic hyperparameters tuning. These enhancements minimize redundant computational costs and improve fault tolerance during deep learning workflow training. Couler is extensively deployed in real-world production scenarios at Ant Group, handling approximately 22k workflows daily, and has successfully improved the CPU/Memory utilization by more than 15% and the workflow completion rate by around 17%.

ASPEN: High-Throughput LoRA Fine-Tuning of Large Language Models with a Single GPU

Transformer-based large language models (LLMs) have demonstrated outstanding performance across diverse domains, particularly when fine-turned for specific domains. Recent studies suggest that the resources required for fine-tuning LLMs can be economized through parameter-efficient methods such as Low-Rank Adaptation (LoRA). While LoRA effectively reduces computational burdens and resource demands, it currently supports only a single-job fine-tuning setup. In this paper, we present ASPEN, a high-throughput framework for fine-tuning LLMs. ASPEN efficiently trains multiple jobs on a single GPU using the LoRA method, leveraging shared pre-trained model and adaptive scheduling. ASPEN is compatible with transformer-based language models like LLaMA and ChatGLM, etc. Experiments show that ASPEN saves 53% of GPU memory when training multiple LLaMA-7B models on NVIDIA A100 80GB GPU and boosts training throughput by about 17% compared to existing methods when training with various pre-trained models on different GPUs. The adaptive scheduling algorithm reduces turnaround time by 24%, end-to-end training latency by 12%, prioritizing jobs and preventing out-of-memory issues.

DLRover: An Elastic Deep Training Extension with Auto Job Resource Recommendation

The cloud is still a popular platform for distributed deep learning (DL) training jobs since resource sharing in the cloud can improve resource utilization and reduce overall costs. However, such sharing also brings multiple challenges for DL training jobs, e.g., high-priority jobs could impact, even interrupt, low-priority jobs. Meanwhile, most existing distributed DL training systems require users to configure the resources (i.e., the number of nodes and resources like CPU and memory allocated to each node) of jobs manually before job submission and can not adjust the job's resources during the runtime. The resource configuration of a job deeply affect this job's performance (e.g., training throughput, resource utilization, and completion rate). However, this usually leads to poor performance of jobs since users fail to provide optimal resource configuration in most cases. \system~is a distributed DL framework can auto-configure a DL job's initial resources and dynamically tune the job's resources to win the better performance. With elastic capability, \system~can effectively adjusts the resources of a job when there are performance issues detected or a job fails because of faults or eviction. Evaluations results show \system~can outperform manual well-tuned resource configurations. Furthermore, in the production Kubernetes cluster of \company, \system~reduces the medium of job completion time by 31\%, and improves the job completion rate by 6\%, CPU utilization by 15\%, and memory utilization by 20\% compared with manual configuration.

SQLFlow: A Bridge between SQL and Machine Learning

Industrial AI systems are mostly end-to-end machine learning (ML) workflows. A typical recommendation or business intelligence system includes many online micro-services and offline jobs. We describe SQLFlow for developing such workflows efficiently in SQL. SQL enables developers to write short programs focusing on the purpose (what) and ignoring the procedure (how). Previous database systems extended their SQL dialect to support ML. SQLFlow (https://sqlflow.org/sqlflow ) takes another strategy to work as a bridge over various database systems, including MySQL, Apache Hive, and Alibaba MaxCompute, and ML engines like TensorFlow, XGBoost, and scikit-learn. We extended SQL syntax carefully to make the extension working with various SQL dialects. We implement the extension by inventing a collaborative parsing algorithm. SQLFlow is efficient and expressive to a wide variety of ML techniques -- supervised and unsupervised learning; deep networks and tree models; visual model explanation in addition to training and prediction; data processing and feature extraction in addition to ML. SQLFlow compiles a SQL program into a Kubernetes-native workflow for fault-tolerable execution and on-cloud deployment. Current industrial users include Ant Financial, DiDi, and Alibaba Group.