Frontier is a new discrete-event simulator for disaggregated LLM serving that incorporates co-location, PDD, AFD, and optimizations, achieving under 4% throughput error and large reductions in latency prediction error versus prior simulators.
hub Canonical reference
SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills
Canonical reference. 86% of citing Pith papers cite this work as background.
31
Pith papers citing it
Background
86% of classified citations
abstract
Large Language Model (LLM) inference consists of two distinct phases - prefill phase which processes the input prompt and decode phase which generates output tokens autoregressively. While the prefill phase effectively saturates GPU compute at small batch sizes, the decode phase results in low compute utilization as it generates one token at a time per request. The varying prefill and decode times also lead to imbalance across micro-batches when using pipeline parallelism, resulting in further inefficiency due to bubbles. We present SARATHI to address these challenges. SARATHI employs chunked-prefills, which splits a prefill request into equal sized chunks, and decode-maximal batching, which constructs a batch using a single prefill chunk and populates the remaining slots with decodes. During inference, the prefill chunk saturates GPU compute, while the decode requests 'piggyback' and cost up to an order of magnitude less compared to a decode-only batch. Chunked-prefills allows constructing multiple decode-maximal batches from a single prefill request, maximizing coverage of decodes that can piggyback. Furthermore, the uniform compute design of these batches ameliorates the imbalance between micro-batches, significantly reducing pipeline bubbles. Our techniques yield significant improvements in inference performance across models and hardware. For the LLaMA-13B model on A6000 GPU, SARATHI improves decode throughput by up to 10x, and accelerates end-to-end throughput by up to 1.33x. For LLaMa-33B on A100 GPU, we achieve 1.25x higher end-to-end-throughput and up to 4.25x higher decode throughput. When used with pipeline parallelism on GPT-3, SARATHI reduces bubbles by 6.29x, resulting in an end-to-end throughput improvement of 1.91x.
hub tools
citation-role summary
background 6
baseline 1
citation-polarity summary
representative citing papers
KVBuffer reduces linear attention decoding latency by up to 45% and increases speculative decoding throughput 5x by buffering keys/values for flexible chunked and parallel computation.
Power capping is illusory in LLM decode as memory-bound operation leaves power headroom untouched on 700 W GPUs, while SM clock locking saves up to 32% energy and three DVFS classes appear across attention types.
MoE-Prefill achieves 1.35-1.59x higher throughput for prefill-only MoE serving by using asynchronous expert parallelism to overlap weight AllGather with computation and prefix-aware routing with true-FLOPs tracking.
GhostServe applies erasure coding to KV cache in host memory for fast recovery from failures in LLM serving, cutting checkpointing latency up to 2.7x and recovery latency 2.1x versus prior methods.
PrefixWall mitigates APC side channels in multi-tenant LLM systems via selective prefix isolation, delivering up to 70% higher cache reuse and 30% lower latency than full-isolation baselines.
SnapStream deploys sparse KV attention in a production inference system on dataflow accelerators, delivering 4x on-chip memory savings for DeepSeek-671B at 128k context with up to 1832 tokens/sec and minimal accuracy loss on LongBench-v2, AIME24, and LiveCodeBench.
DuoAttention identifies retrieval heads requiring full KV cache and streaming heads using constant-length cache to reduce memory and latency in long-context LLM inference.
A training-inference consistent segmented execution framework for long-context LLMs matches full-context performance with substantially lower peak memory at very long lengths.
SPECTRE achieves up to 2.28x speedup for large-model LLM serving by running speculative draft generation and target verification in parallel using idle tail-model services.
SPIN co-designs sparse attention with hierarchical memory to achieve 1.66-5.66x higher throughput, 7-9x lower TTFT, and up to 58% lower TPOT than vLLM and original sparse implementations.
AMMA is a memory-centric multi-chiplet architecture using HBM-PNM cubes, custom logic dies, hybrid parallelism, and reordered collectives that delivers 15.5X lower attention latency and 6.9X lower energy than NVIDIA H100 for 1M context serving.
Salca is a new ASIC accelerator that achieves 3.82× speedup and 74.19× energy efficiency over A100 for long-context attention via dual-compression dynamic sparse attention and pipelined hardware.
NPUMoE accelerates MoE LLM inference on Apple Silicon NPUs via offline-calibrated static expert tiers, grouped execution, and load-aware graph residency, delivering 1.32x-5.55x lower latency and 1.81x-7.37x better energy efficiency.
MARS coordinates heterogeneous GPU-CPU resources for agentic LLM workloads via decoupled admission control and agent-centric KV cache management, delivering up to 5.94x lower latency and 1.87x faster task completion.
A flow-control framework for LLM inference derives necessary and sufficient stability conditions and experimentally improves throughput, latency, and KV cache stability over common baselines.
Valve jointly bounds preemption latency and rate for online-offline LLM colocation on GPUs, delivering 34.6% higher cluster utilization and a 2,170-GPU saving in a production deployment of 8,054 GPUs with under 5% TTFT and 2% TPOT impact.
RetroInfer introduces the wave index and wave buffer to realize sparse KV-cache attention for long-context LLM inference with up to 4.4X throughput gains while matching full-attention accuracy.
The paper develops fluid-guided online scheduling algorithms (WAIT and Nested WAIT) for LLM inference that handle endogenous KV-cache memory growth and improve stability and latency over baselines in simulations.
BatchLLM achieves 1.3x-10.8x higher throughput than vLLM and SGLang for batched LLM inference with prefix sharing via global prefix identification, decoding-first reordering, and memory-centric token batching.
HybridFlow combines single- and multi-controller paradigms with a 3D-HybridEngine to deliver 1.53x to 20.57x higher throughput for various RLHF algorithms compared to prior systems.
AlignedServe uses prefix-aware batching, large CPU in-flight request pools, batch scheduling, and GPU-to-GPU KV prefetching to raise decoding throughput up to 1.98x and cut latency up to 7.4x versus prior serving systems.
Reasoning workloads shift LLM inference to a capacity-bound regime where KV-cache fragmentation limits data parallelism, tensor parallelism unlocks memory at the 32B scale, and MoE models require hybrid strategies to avoid routing latency.
Empirical study finds non-linear, model-size-dependent throughput degradation from offloading and high model-state reload costs from preemption in multi-LLM serving.
citing papers explorer
-
SnapStream: Efficient Long Sequence Decoding on Dataflow Accelerators
SnapStream deploys sparse KV attention in a production inference system on dataflow accelerators, delivering 4x on-chip memory savings for DeepSeek-671B at 128k context with up to 1832 tokens/sec and minimal accuracy loss on LongBench-v2, AIME24, and LiveCodeBench.
-
RetroInfer: A Vector Storage Engine for Scalable Long-Context LLM Inference
RetroInfer introduces the wave index and wave buffer to realize sparse KV-cache attention for long-context LLM inference with up to 4.4X throughput gains while matching full-attention accuracy.
-
Optimizing LLM Inference: Fluid-Guided Online Scheduling with Memory Constraints
The paper develops fluid-guided online scheduling algorithms (WAIT and Nested WAIT) for LLM inference that handle endogenous KV-cache memory growth and improve stability and latency over baselines in simulations.
-
ODMA: On-Demand Memory Allocation Strategy for LLM Serving on LPDDR-Class Accelerators
ODMA raises KV-cache utilization by up to 19.25% and throughput by 23-27% on Cambricon MLU accelerators by dynamically adjusting prediction buckets and using a safety pool for LLM serving.