Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve

• URL: http://arxiv.org/abs/2403.02310v3
• Date: Mon, 17 Jun 2024 21:10:46 GMT
• Title: Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve
• Authors: Amey Agrawal, Nitin Kedia, Ashish Panwar, Jayashree Mohan, Nipun Kwatra, Bhargav S. Gulavani, Alexey Tumanov, Ramachandran Ramjee,
• Abstract summary: We introduce an efficient inference scheduler, Sarathi-Serve, to address the tradeoffs between high throughput and low latency. Our techniques yield significant improvements in inference performance across models and hardware under tail latency.
• Score: 9.854130239429487
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Each LLM serving request goes through two phases. The first is prefill which processes the entire input prompt and produces the first output token and the second is decode which generates the rest of output tokens, one-at-a-time. Prefill iterations have high latency but saturate GPU compute due to parallel processing of the input prompt. In contrast, decode iterations have low latency but also low compute utilization because a decode iteration processes only a single token per request. This makes batching highly effective for decodes and consequently for overall throughput. However, batching multiple requests leads to an interleaving of prefill and decode iterations which makes it challenging to achieve both high throughput and low latency. We introduce an efficient LLM inference scheduler, Sarathi-Serve, to address this throughput-latency tradeoff. Sarathi-Serve introduces chunked-prefills which splits a prefill request into near equal sized chunks and creates stall-free schedules that adds new requests in a batch without pausing ongoing decodes. Stall-free scheduling unlocks the opportunity to improve throughput with large batch sizes while minimizing the effect of batching on latency. Furthermore, uniform batches in Sarathi-Serve ameliorate the imbalance between iterations resulting in minimal pipeline bubbles. Our techniques yield significant improvements in inference performance across models and hardware under tail latency constraints. For Mistral-7B on single A100 GPUs, we achieve 2.6x higher serving capacity and up to 3.7x higher serving capacity for the Yi-34B model on two A100 GPUs as compared to vLLM. When used with pipeline parallelism on Falcon-180B, Sarathi-Serve provides up to 5.6x gain in the end-to-end serving capacity. The source code for Sarathi-Serve is available at https://github.com/microsoft/sarathi-serve.

Related papers

• POD-Attention: Unlocking Full Prefill-Decode Overlap for Faster LLM Inference [9.164093249308419]
  We present POD-Attention -- the first GPU kernel that efficiently computes attention for hybrid batches. POD-Attention aims to maximize the utilization of both compute and memory bandwidth by carefully allocating the GPU's resources.
  arXiv  Detail & Related papers  (2024-10-23T17:06:56Z)
• MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models [58.3342517278868]
  This paper describes the design of Mixed-precision AutoRegressive LINear kernels. It shows that batchsizes up to 16-32 can be supported with close to maximum ($4times$) quantization speedup. MarLIN accomplishes this via a combination of techniques, such as asynchronous memory access, complex task scheduling and pipelining.
  arXiv  Detail & Related papers  (2024-08-21T16:10:41Z)
• MInference 1.0: Accelerating Pre-filling for Long-Context LLMs via Dynamic Sparse Attention [36.49445805074941]
  MInference (Milliontokens Inference) is a sparse calculation method designed to accelerate pre-filling of long-sequence processing. We demonstrate that MInference effectively reduces inference latency by up to 10x for pre-filling on an A100, while maintaining accuracy.
  arXiv  Detail & Related papers  (2024-07-02T17:59:56Z)
• Hardware-Aware Parallel Prompt Decoding for Memory-Efficient Acceleration of LLM Inference [19.167604927651073]
  Auto-regressive decoding of Large Language Models (LLMs) results in significant overheads in their hardware performance. We propose a novel parallel prompt decoding that requires only $0.0002$% trainable parameters, enabling efficient training on a single A100-40GB GPU in just 16 hours. Our approach demonstrates up to 2.49$times$ speedup and maintains a minimal memory overhead of just $0.0004$%.
  arXiv  Detail & Related papers  (2024-05-28T22:19:30Z)
• QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving [52.31791050376249]
  Quantization can accelerate large language model (LLM) inference. Existing INT4 quantization methods suffer from significant runtime overhead when dequantizing weights or partial sums. We introduce QoQ, a W4A8KV4 quantization algorithm with 4-bit weight, 8-bit activation, and 4-bit KV cache. QServe improves the maximum achievable serving of Llama-3-8B by 1.2x on A100, 1.4x on L40S; and Qwen-721.5B by 2.4x on A100, 3.5x on L40S.
  arXiv  Detail & Related papers  (2024-05-07T17:59:30Z)
• HiRE: High Recall Approximate Top-$k$ Estimation for Efficient LLM Inference [68.59839755875252]
  HiRE comprises of two novel components: (i) a compression scheme to cheaply predict top-$k$ rows/columns with high recall, followed by full computation restricted to the predicted subset, and (ii) DA-TOP-$k$: an efficient multi-device approximate top-$k$ operator. We demonstrate that on a one billion parameter model, HiRE applied to both the softmax as well as feedforward layers, achieves almost matching pretraining and downstream accuracy, and speeds up inference latency by $1.47times$ on a single TPUv5e device.
  arXiv  Detail & Related papers  (2024-02-14T18:04:36Z)
• SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills [9.821549185732199]
  Large Language Model (LLM) inference consists of two distinct phases - prefill and decode. decode phase results in low compute utilization as it generates one token at a time per request. Chunked-prefills allows constructing multiple decode-maximal batches from a single prefill request. Our techniques yield significant improvements in inference performance across models and hardware.
  arXiv  Detail & Related papers  (2023-08-31T00:03:02Z)
• SkipDecode: Autoregressive Skip Decoding with Batching and Caching for Efficient LLM Inference [17.947904697850433]
  We present SkipDecode, a token-level early exit method for batch inferencing and KeyValue caching. It overcomes prior constraints by setting up singular-level exit point for every token in a batch at each sequence position. It also guarantees a monotonic decrease in exit points, thereby eliminating the need to recompute KV Caches for preceding tokens.
  arXiv  Detail & Related papers  (2023-07-05T19:59:09Z)
• SqueezeLLM: Dense-and-Sparse Quantization [80.32162537942138]
  Main bottleneck for generative inference with LLMs is memory bandwidth, rather than compute, for single batch inference. We introduce SqueezeLLM, a post-training quantization framework that enables lossless compression to ultra-low precisions of up to 3-bit. Our framework incorporates two novel ideas: (i) sensitivity-based non-uniform quantization, which searches for the optimal bit precision assignment based on second-order information; and (ii) the Dense-and-Sparse decomposition that stores outliers and sensitive weight values in an efficient sparse format.
  arXiv  Detail & Related papers  (2023-06-13T08:57:54Z)
• RSC: Accelerating Graph Neural Networks Training via Randomized Sparse Computations [56.59168541623729]
  Training graph neural networks (GNNs) is time consuming because sparse graph-based operations are hard to be accelerated by hardware. We explore trading off the computational precision to reduce the time complexity via sampling-based approximation. We propose Randomized Sparse Computation, which for the first time demonstrate the potential of training GNNs with approximated operations.
  arXiv  Detail & Related papers  (2022-10-19T17:25:33Z)
• Parallelising the Queries in Bucket Brigade Quantum RAM [69.43216268165402]
  Quantum algorithms often use quantum RAMs (QRAM) for accessing information stored in a database-like manner. We show a systematic method to significantly reduce the effective query time by using Clifford+T gate parallelism. We conclude that, in theory, fault-tolerant bucket brigade quantum RAM queries can be performed approximately with the speed of classical RAM.
  arXiv  Detail & Related papers  (2020-02-21T14:50:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.