KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantization

LLMs are seeing growing use for applications such as document analysis and summarization which require large context windows, and with these large context windows KV cache activations surface as the dominant contributor to memory consumption during inference. Quantization is a promising approach for compressing KV cache activations; however, existing solutions fail to represent activations accurately in ultra-low precisions, such as sub-4-bit. In this work, we present KVQuant, which addresses this problem by incorporating novel methods for quantizing cached KV activations, including: (i) Per-Channel Key Quantization, where we adjust the dimension along which we quantize the Key activations to better match the distribution; (ii) Pre-RoPE Key Quantization, where we quantize Key activations before the rotary positional embedding to mitigate its impact on quantization; (iii) Non-Uniform KV Cache Quantization, where we derive per-layer sensitivity-weighted non-uniform datatypes that better represent the distributions; (iv) Per-Vector Dense-and-Sparse Quantization, where we isolate outliers separately for each vector to minimize skews in quantization ranges; and (v) Q-Norm, where we normalize quantization centroids in order to mitigate distribution shift, providing additional benefits for 2-bit quantization. By applying our method to the LLaMA, LLaMA-2, and Mistral models, we achieve $<0.1$ perplexity degradation with 3-bit quantization on both Wikitext-2 and C4, outperforming existing approaches. Our method enables serving the LLaMA-7B model with a context length of up to 1 million on a single A100-80GB GPU and up to 10 million on an 8-GPU system.

MoCA: Memory-Centric, Adaptive Execution for Multi-Tenant Deep Neural Networks

Driven by the wide adoption of deep neural networks (DNNs) across different application domains, multi-tenancy execution, where multiple DNNs are deployed simultaneously on the same hardware, has been proposed to satisfy the latency requirements of different applications while improving the overall system utilization. However, multi-tenancy execution could lead to undesired system-level resource contention, causing quality-of-service (QoS) degradation for latency-critical applications. To address this challenge, we propose MoCA, an adaptive multi-tenancy system for DNN accelerators. Unlike existing solutions that focus on compute resource partition, MoCA dynamically manages shared memory resources of co-located applications to meet their QoS targets. Specifically, MoCA leverages the regularities in both DNN operators and accelerators to dynamically modulate memory access rates based on their latency targets and user-defined priorities so that co-located applications get the resources they demand without significantly starving their co-runners. We demonstrate that MoCA improves the satisfaction rate of the service level agreement (SLA) up to 3.9x (1.8x average), system throughput by 2.3x (1.7x average), and fairness by 1.3x (1.2x average), compared to prior work.

Full Stack Optimization of Transformer Inference: a Survey

Recent advances in state-of-the-art DNN architecture design have been moving toward Transformer models. These models achieve superior accuracy across a wide range of applications. This trend has been consistent over the past several years since Transformer models were originally introduced. However, the amount of compute and bandwidth required for inference of recent Transformer models is growing at a significant rate, and this has made their deployment in latency-sensitive applications challenging. As such, there has been an increased focus on making Transformer models more efficient, with methods that range from changing the architecture design, all the way to developing dedicated domain-specific accelerators. In this work, we survey different approaches for efficient Transformer inference, including: (i) analysis and profiling of the bottlenecks in existing Transformer architectures and their similarities and differences with previous convolutional models; (ii) implications of Transformer architecture on hardware, including the impact of non-linear operations such as Layer Normalization, Softmax, and GELU, as well as linear operations, on hardware design; (iii) approaches for optimizing a fixed Transformer architecture; (iv) challenges in finding the right mapping and scheduling of operations for Transformer models; and (v) approaches for optimizing Transformer models by adapting the architecture using neural architecture search. Finally, we perform a case study by applying the surveyed optimizations on Gemmini, the open-source, full-stack DNN accelerator generator, and we show how each of these approaches can yield improvements, compared to previous benchmark results on Gemmini. Among other things, we find that a full-stack co-design approach with the aforementioned methods can result in up to 88.7x speedup with a minimal performance degradation for Transformer inference.

CoSA: Scheduling by Constrained Optimization for Spatial Accelerators

Recent advances in Deep Neural Networks (DNNs) have led to active development of specialized DNN accelerators, many of which feature a large number of processing elements laid out spatially, together with a multi-level memory hierarchy and flexible interconnect. While DNN accelerators can take advantage of data reuse and achieve high peak throughput, they also expose a large number of runtime parameters to the programmers who need to explicitly manage how computation is scheduled both spatially and temporally. In fact, different scheduling choices can lead to wide variations in performance and efficiency, motivating the need for a fast and efficient search strategy to navigate the vast scheduling space. To address this challenge, we present CoSA, a constrained-optimization-based approach for scheduling DNN accelerators. As opposed to existing approaches that either rely on designers' heuristics or iterative methods to navigate the search space, CoSA expresses scheduling decisions as a constrained-optimization problem that can be deterministically solved using mathematical optimization techniques. Specifically, CoSA leverages the regularities in DNN operators and hardware to formulate the DNN scheduling space into a mixed-integer programming (MIP) problem with algorithmic and architectural constraints, which can be solved to automatically generate a highly efficient schedule in one shot. We demonstrate that CoSA-generated schedules significantly outperform state-of-the-art approaches by a geometric mean of up to 2.5x across a wide range of DNN networks while improving the time-to-solution by 90x.

Efficient emotion recognition using hyperdimensional computing with combinatorial channel encoding and cellular automata

In this paper, a hardware-optimized approach to emotion recognition based on the efficient brain-inspired hyperdimensional computing (HDC) paradigm is proposed. Emotion recognition provides valuable information for human-computer interactions, however the large number of input channels (>200) and modalities (>3) involved in emotion recognition are significantly expensive from a memory perspective. To address this, methods for memory reduction and optimization are proposed, including a novel approach that takes advantage of the combinatorial nature of the encoding process, and an elementary cellular automaton. HDC with early sensor fusion is implemented alongside the proposed techniques achieving two-class multi-modal classification accuracies of >76% for valence and >73% for arousal on the multi-modal AMIGOS and DEAP datasets, almost always better than state of the art. The required vector storage is seamlessly reduced by 98% and the frequency of vector requests by at least 1/5. The results demonstrate the potential of efficient hyperdimensional computing for low-power, multi-channeled emotion recognition tasks.

Gemmini: An Agile Systolic Array Generator Enabling Systematic Evaluations of Deep-Learning Architectures

Advances in deep learning and neural networks have resulted in the rapid development of hardware accelerators that support them. A large majority of ASIC accelerators, however, target a single hardware design point to accelerate the main computational kernels of deep neural networks such as convolutions or matrix multiplication. On the other hand, the spectrum of use-cases for neural network accelerators, ranging from edge devices to cloud, presents a prime opportunity for agile hardware design and generator methodologies. We present Gemmini -- an open source and agile systolic array generator enabling systematic evaluations of deep-learning architectures. Gemmini generates a custom ASIC accelerator for matrix multiplication based on a systolic array architecture, complete with additional functions for neural network inference. Gemmini runs with the RISC-V ISA, and is integrated with the Rocket Chip System-on-Chip generator ecosystem, including Rocket in-order cores and BOOM out-of-order cores. Through an elaborate design space exploration case study, this work demonstrates the selection processes of various parameters for the use-case of inference on edge devices. Selected design points achieve two to three orders of magnitude speedup in deep neural network inference compared to the baseline execution on a host processor. Gemmini-generated accelerators were used in the fabrication of test systems-on-chip in TSMC 16nm and Intel 22FFL process technologies.