AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

DuetServe: Harmonizing Prefill and Decode for LLM Serving via Adaptive GPU Multiplexing

Lei Gao, Chaoyi Jiang, Hossein Entezari Zarch, Daniel Wong, Mark D. Hill, Murali Annavaram

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Distributed LLM inference systems · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use DuetServe adaptive SM multiplexing instead of vLLM (short output, OSL=64)vLLM · Synthetic ISL=4096 OSL=64throughput (req/s): 10.0 -> 12.8Improved
use DuetServe instead of vLLM (OSL=1024)vLLM · Synthetic ISL=4096 OSL=1024throughput (req/s): 8.8 -> 9.8Improved
use DuetServe instead of vLLM (decode-heavy, OSL=2048)vLLM · Synthetic ISL=4096 OSL=2048throughput (req/s): 6.9 -> 7.2No clear effect
use DuetServe instead of SGLang-DefaultSGLang-Default · Azure-Code QPS=16output request throughput (req/s): 12.43 -> 13.57Improved
use DuetServe instead of vLLM at QPS=5vLLM · Mooncake Conversationoutput request throughput (req/s): 2.14 -> 2.72Improved
use adaptive SM partitioning instead of PD disaggregation (Dynamo, 8 GPUs)Dynamo · Azure-Conv Qwen3-32B QPS=24throughput (req/s): 5.69 -> 8.02Improved
use adaptive SM partitioning instead of PD disaggregation (Dynamo, 8 GPUs)Dynamo · Azure-Conv Qwen3-32B QPS=24TTFT (s): 110.2 -> 58.9Improved
use adaptive SM partitioning instead of PD disaggregation (Dynamo, 8 GPUs)Dynamo · Azure-Conv Qwen3-32B QPS=24TBT (ms): 23.1 -> 104.7Hurt
use adaptive SM partitioning instead of PD disaggregation (Dynamo, 8 GPUs)Dynamo · Azure-Conv Qwen3-32B QPS=24GPU utilization (%): 74.6% -> 93.5%Improved
use adaptive scheduling instead of static SM partitions (Sd22-Sp44, Sd33-Sp33, Sd44-Sp22) (ablation)DuetServe · Azure-Code/Azure-Conv/Mooncake, Qwen3-8B/14BthroughputImproved
use a more conservative (overestimating) roofline decode latency model (ablation)DuetServe · Roofline model, small TPC decodepartition qualityNo clear effect
add multi-GPU TP2 DuetServe vs Dynamo-1P1DDynamo-1P1D · Azure-Code TP=2throughputImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy