PyTorchKR Papers

업데이트된 논문

총 4편
태그·기간 등 상세 필터

카테고리 탐색

arXiv:2605.22781v3cs.OS

DeltaBox: Millisecond-Scale Sandbox Checkpoint and Rollback for Stateful AI Agents

Emerging LLM-powered AI agent workloads, such as test-time tree search and reinforcement learning (RL), require high-frequency state exploration, relying on rapid checkpoint/restore (C/R) of the complete sandbox state, including files and process state. Existing mechanisms duplicate the entire state, causing hundreds of milliseconds to seconds of latency per C/R, which severely bottlenecks deep search and large-scale fan-outs. We observe that consecutive sandbox states between checkpoints differ only marginally. Therefore, instead of full duplication, a sandbox should only duplicate the changes between consecutive checkpoints (Key Insight), yet the lack of OS support makes this hard to realize. We propose a new OS-level abstraction, DeltaState, that delivers change-based C/R for agents through co-designed OS mechanisms. First, DeltaFS checkpoints file state by freezing the writable layer and inserting a new one, turning updates into copy-on-write and rollback into a layer switch. Second, DeltaCR quickly restores process state by forking from a frozen template process. A coordinator pairs the two so each checkpoint is one consistent (filesystem, memory) state. Building on these mechanisms, we present DeltaBox, an agent sandbox with millisecond-level C/R. On SWE-bench, DeltaBox hides its checkpoint work (10.83 ms) under inference and rolls back in a mean 1.86 ms, with the same warm-template fork driving low-cost RL rollout fan-out. DeltaBox is open-sourced at https://github.com/delta-box.

arXiv:2611.10453v2cs.AI

BunshinRL: Whole-Workspace Forking for Agentic Reinforcement Learning Rollouts

Agentic reinforcement learning (RL) rollouts interleave model reasoning with tool calls in persistent workspaces. Scaling these rollouts incurs substantial memory costs for concurrent environments, while slow tool calls delay entire rollout batches. We present BunshinRL, an agentic RL rollout system that exploits shared environment state and recurring tool calls through three techniques: memory sharing across environments, tool-prefix caching, and speculative tool execution. These techniques build on BunshinFork, a novel live whole-workspace forking mechanism for containers. BunshinFork reconstructs process trees, shares memory copy-on-write, and preserves file, communication, and GUI state, enabling cached-state reuse and rollback after incorrect speculation. We implement BunshinRL and BunshinFork through Linux kernel module and CRIU modifications. BunshinRL reduces peak memory by up to 43%, total memory consumption by up to 51%, and end-to-end rollout latency by 23%.

arXiv:2604.26963v3cs.OS

MARS: Towards High Goodput and Low KV Thrashing in Agent Serving under GPU Memory Constraints

Self-hosted LLM agents offer greater control over sensitive data and model customization while avoiding per-request API fees. However, serving concurrent, multi-turn workflows within limited GPU memory makes it difficult to achieve high workflow-level goodput. In particular, as workflows accumulate context, their reusable key-value (KV) caches can easily outgrow GPU capacity, triggering repeated eviction and recovery, known as KV cache thrashing. This process consumes resources and delays workflow completion. Although existing approaches propose various methods, our motivation study shows that the evaluated baselines still incur substantial thrashing overhead or fail to address the workflow-level goodput issue. We present MARS, an agent serving system that coordinates workflow admission and in-engine scheduling to achieve high workflow-level goodput on memory-constrained devices while avoiding KV cache thrashing. Specifically, outside the engine, an admission controller selects root workflows using initial prefix footprint estimates and live KV availability, adapting admissions to memory pressure. Inside, an agent-centric scheduler prioritizes turn completion and coordinates prefill allocation with KV reclamation to reduce disruptive evictions. Across two models and three GPU configurations, MARS achieves 1.59x-7.56x mean end-to-end latency speedups over the fastest baseline. Integrated as the serving backend for OpenCode, Codex, and OpenHands, MARS further improves mean workflow completion time by 1.17x-1.87x. Our source code is publicly available at https://github.com/Afterglow231/MARS_preview .

arXiv:2610.38196v2cs.DC

GearUp: DAG-aware Acceleration for Serverless Inference Serving via Slack-guided Pre-loading

Serverless platforms increasingly host machine learning (ML) inference workflows, modeled as directed acyclic graphs (DAGs) of interdependent functions. Beyond container startup, each ML node must load libraries, model weights, and CUDA kernels before it can run. These costs accumulate along dependent DAG paths and place initialization on the end-to-end critical path. Existing systems optimize one half of this problem. DAG-agnostic pre-loading prepares nodes the request may never reach, wasting host and GPU memory. DAG-aware serving routes and scales whole functions, leaving the loading stages inside a node untouched. This paper presents GearUp, a serverless DAG inference serving system that removes ML-node loading from the end-to-end critical path at low resource cost. GearUp profiles the loading stages of each node and schedules them inside the slack time the DAG exposes, hiding initialization behind upstream execution. It reuses initialized instances within and across requests to avoid redundant cold starts. A probabilistic, traffic-adaptive keep-alive policy sets how long idle instances are retained, balancing memory cost against cold-start risk under bursty load. On the AMBER Alert and FINRA-ML workflows with NVIDIA L40S GPUs, GearUp reduces median end-to-end latency by 64.5% and 75.9% against SMIless, for median speedups of 3.1x and 10.0x over cold execution. Integrated GPU and host memory allocation fall by up to 83.9% and 94.6% against eager pre-loading in InstaInfer. GearUp achieves 4.2x strong scaling on four workers.