VibeServe demonstrates that AI agents can synthesize bespoke LLM serving systems end-to-end, remaining competitive with vLLM in standard settings while outperforming it in six non-standard scenarios involving unusual models, workloads, or hardware.
hub Canonical reference
Gonzalez, and Ion Stoica
Canonical reference. 83% of citing Pith papers cite this work as background.
22
Pith papers citing it
1
external citations · Pith
Background
83% of classified citations
abstract
Large language model (LLM) applications are evolving beyond simple chatbots into dynamic, general-purpose agentic programs, which scale LLM calls and output tokens to help AI agents reason, explore, and solve complex tasks. However, existing LLM serving systems ignore dependencies between programs and calls, missing significant opportunities for optimization. Our analysis reveals that programs submitted to LLM serving engines experience long cumulative wait times, primarily due to head-of-line blocking at both the individual LLM request and the program. To address this, we introduce Autellix, an LLM serving system that treats programs as first-class citizens to minimize their end-to-end latencies. Autellix intercepts LLM calls submitted by programs, enriching schedulers with program-level context. We propose two scheduling algorithms-for single-threaded and distributed programs-that preempt and prioritize LLM calls based on their programs' previously completed calls. Our evaluation demonstrates that across diverse LLMs and agentic workloads, Autellix improves throughput of programs by 4-15x at the same latency compared to state-of-the-art systems, such as vLLM.
hub tools
citation-role summary
background 5
other 1
citation-polarity summary
representative citing papers
SmoothAgent introduces lookahead context engineering to eliminate transformation overhead in LLM agents, reducing TTFT by up to 11.9x through proactive KV cache preparation.
Presents a distribution-aware scheduling framework for LLM inference that reduces P99 TTLT by 35-50% and TTFT by 34-47% versus SRPT with perfect length knowledge using statistical signals instead of predictions.
AGENTSERVESIM is a new simulator that models multi-turn LLM agent serving at program granularity and reproduces real-system performance within 6% error on commodity CPUs.
FATE reduces normalized makespan and P95 latency in real LLM workflow DAGs to 0.675 and 0.677 by jointly preserving multiple future execution states, outperforming RoundRobin by 32.5% and the strongest baseline by 8.9%.
SAGA introduces workflow-atomic scheduling for compound AI agents, achieving 1.64x lower task completion time and 1.22x better memory utilization than vLLM on a 64-GPU cluster at the cost of 30% lower peak throughput.
This work delivers the first measurements of performance-energy trade-offs across four multi-request LLM workflow patterns on A100 GPUs using vLLM and Parrot.
Balancing only first-turn agent requests while keeping follow-ups cache-aware raises LLM cluster TPS 10-16% (colocation) and prefill TPS 2-34% (disaggregation) over SOTA schedulers on production agent traces.
MORI improves throughput 20-71% and TTFT 18-43% over baselines by ranking programs on a continuous idleness spectrum and shifting the GPU-CPU boundary to match capacity in agentic LLM serving.
A four-primitive agent runtime layer between agent frameworks and LLM serving engines raises cross-session KV-cache hit rate by 13–37 percentage points in the paper's experiments.
KAIROS reduces power by 27% on average (up to 39.8%) for agentic AI inference by using long-lived context to jointly manage GPU frequency, concurrency, and request routing across instances.
Scepsy schedules arbitrary multi-LLM agentic workflows on GPU clusters by constructing Aggregate LLM Pipelines from stable per-LLM execution time shares, then searching fractional GPU allocations, tensor parallelism, and replica counts to achieve up to 2.4x higher throughput and 27x lower latency.
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.
R-DSGD and R-DSGD-M under (δ,c)-robust aggregation have tight Byzantine error floors under (B,ζ)-bounded dissimilarity; local momentum eliminates the stochastic-noise term but not the heterogeneity term.
ForkKV uses copy-on-write disaggregated KV cache with DualRadixTree and ResidualAttention kernels to deliver up to 3x throughput over prior multi-LoRA serving systems with negligible quality loss.
TokenDance scales multi-agent LLM serving to 2.7x more concurrent agents by collective KV cache reuse and block-sparse diff encoding that achieves 11-17x compression.
TTL-based KV-cache pinning that uses predicted tool-call durations and queueing-delay costs cuts agent job completion time by up to 8x.
TokenCake introduces agent-aware temporal and spatial schedulers for KV cache management in LLM multi-agent serving, claiming over 47% lower end-to-end latency and up to 16.9% better GPU memory utilization than vLLM on representative benchmarks.
AsymCache combines Multi-Segment Attention, position-aware eviction, and adaptive chunking to cut TTFT by up to 2.03x and TPOT by up to 1.71x versus recent baselines in LLM serving.
PBKV predicts agent invocations in dynamic LLM workflows to manage KV-cache reuse, delivering up to 1.85x speedup over LRU and 1.26x over KVFlow.
AgentOpt introduces a framework-agnostic package that uses algorithms like UCB-E to find cost-effective model assignments in multi-step LLM agent pipelines, cutting evaluation budgets by 62-76% while maintaining near-optimal accuracy on benchmarks.
Introduces a benchmarking suite with common workload adapters, event schemas, and an evidence gate connecting WebArena Verified, SWE-Gym, and MiniWoB++ for tool-using agents.
citing papers explorer
-
VibeServe: Can AI Agents Build Bespoke LLM Serving Systems?
VibeServe demonstrates that AI agents can synthesize bespoke LLM serving systems end-to-end, remaining competitive with vLLM in standard settings while outperforming it in six non-standard scenarios involving unusual models, workloads, or hardware.
-
SmoothAgent: Efficient Long-Horizon LLM-Based Agent Serving with Lookahead Context Engineering
SmoothAgent introduces lookahead context engineering to eliminate transformation overhead in LLM agents, reducing TTFT by up to 11.9x through proactive KV cache preparation.
-
Beyond Prediction: Tail-Aware Scheduling for LLM Inference
Presents a distribution-aware scheduling framework for LLM inference that reduces P99 TTLT by 35-50% and TTFT by 34-47% versus SRPT with perfect length knowledge using statistical signals instead of predictions.
-
AGENTSERVESIM: A Hardware-aware Simulator for Multi-Turn LLM Agent Serving
AGENTSERVESIM is a new simulator that models multi-turn LLM agent serving at program granularity and reproduces real-system performance within 6% error on commodity CPUs.
-
FATE: Future-State-Aware Scheduling for Heterogeneous LLM Workflows
FATE reduces normalized makespan and P95 latency in real LLM workflow DAGs to 0.675 and 0.677 by jointly preserving multiple future execution states, outperforming RoundRobin by 32.5% and the strongest baseline by 8.9%.
-
SAGA: Workflow-Atomic Scheduling for AI Agent Inference on GPU Clusters
SAGA introduces workflow-atomic scheduling for compound AI agents, achieving 1.64x lower task completion time and 1.22x better memory utilization than vLLM on a 64-GPU cluster at the cost of 30% lower peak throughput.
-
Characterizing Performance-Energy Trade-offs of Large Language Models in Multi-Request Workflows
This work delivers the first measurements of performance-energy trade-offs across four multi-request LLM workflow patterns on A100 GPUs using vLLM and Parrot.
-
SMetric: Rethink LLM Scheduling for Serving Agents with Balanced Session-centric Scheduling
Balancing only first-turn agent requests while keeping follow-ups cache-aware raises LLM cluster TPS 10-16% (colocation) and prefill TPS 2-34% (disaggregation) over SOTA schedulers on production agent traces.
-
Idleness is Relative: Exploiting Tool-Call Idle Windows for Offloading in Agentic Systems with MORI
MORI improves throughput 20-71% and TTFT 18-43% over baselines by ranking programs on a continuous idleness spectrum and shifting the GPU-CPU boundary to match capacity in agentic LLM serving.
-
A Policy-Driven Runtime Layer for Agentic LLM Serving
A four-primitive agent runtime layer between agent frameworks and LLM serving engines raises cross-session KV-cache hit rate by 13–37 percentage points in the paper's experiments.
-
KAIROS: Stateful, Context-Aware Power-Efficient Agentic Inference Serving
KAIROS reduces power by 27% on average (up to 39.8%) for agentic AI inference by using long-lived context to jointly manage GPU frequency, concurrency, and request routing across instances.
-
Scepsy: Serving Agentic Workflows Using Aggregate LLM Pipelines
Scepsy schedules arbitrary multi-LLM agentic workflows on GPU clusters by constructing Aggregate LLM Pipelines from stable per-LLM execution time shares, then searching fractional GPU allocations, tensor parallelism, and replica counts to achieve up to 2.4x higher throughput and 27x lower latency.
-
MARS: Efficient, Adaptive Co-Scheduling for Heterogeneous Agentic Systems
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.
-
Tessera: Unlocking Heterogeneous GPUs through Kernel-Granularity Disaggregation
R-DSGD and R-DSGD-M under (δ,c)-robust aggregation have tight Byzantine error floors under (B,ζ)-bounded dissimilarity; local momentum eliminates the stochastic-noise term but not the heterogeneity term.
-
ForkKV: Scaling Multi-LoRA Agent Serving via Copy-on-Write Disaggregated KV Cache
ForkKV uses copy-on-write disaggregated KV cache with DualRadixTree and ResidualAttention kernels to deliver up to 3x throughput over prior multi-LoRA serving systems with negligible quality loss.
-
TokenDance: Scaling Multi-Agent LLM Serving via Collective KV Cache Sharing
TokenDance scales multi-agent LLM serving to 2.7x more concurrent agents by collective KV cache reuse and block-sparse diff encoding that achieves 11-17x compression.
-
Continuum: Efficient and Robust Multi-Turn LLM Agent Scheduling with KV Cache Time-to-Live
TTL-based KV-cache pinning that uses predicted tool-call durations and queueing-delay costs cuts agent job completion time by up to 8x.
-
TokenCake: A KV-Cache-centric Serving Framework for LLM-based Multi-Agent Applications
TokenCake introduces agent-aware temporal and spatial schedulers for KV cache management in LLM multi-agent serving, claiming over 47% lower end-to-end latency and up to 16.9% better GPU memory utilization than vLLM on representative benchmarks.
-
Multi-Segment Attention: Enabling Efficient KV-Cache Management for Faster Large Language Model Serving
AsymCache combines Multi-Segment Attention, position-aware eviction, and adaptive chunking to cut TTFT by up to 2.03x and TPOT by up to 1.71x versus recent baselines in LLM serving.
-
Efficient Serving for Dynamic Agent Workflows with Prediction-based KV-Cache Management
PBKV predicts agent invocations in dynamic LLM workflows to manage KV-cache reuse, delivering up to 1.85x speedup over LRU and 1.26x over KVFlow.
-
AgentOpt v0.1 Technical Report: Client-Side Optimization for LLM-Based Agent
AgentOpt introduces a framework-agnostic package that uses algorithms like UCB-E to find cost-effective model assignments in multi-step LLM agent pipelines, cutting evaluation budgets by 62-76% while maintaining near-optimal accuracy on benchmarks.
-
An Executable Benchmarking Suite for Tool-Using Agents
Introduces a benchmarking suite with common workload adapters, event schemas, and an evidence gate connecting WebArena Verified, SWE-Gym, and MiniWoB++ for tool-using agents.