Tag
论文提出 CQ2-CARE 方法，在固定 token 预算下为编码智能体选择完整文件而非压缩或检索片段，从而在 SWE-rebench 上显著提升问题修复率，超过压缩器、检索器及专用训练方法。
The paper introduces Success-Referenced On-Policy Distillation (SR-OPD), which uses a student's own successful rollouts as references to selectively allocate teacher supervision, achieving comparable reasoning performance on math benchmarks using only 3.46–5.02% of the teacher-input tokens required by vanilla on-policy distillation.
The paper proposes LoHi, a training-free framework that jointly allocates frames, resolution, and front-end decoding latency for long-video understanding with VLMs, combining dense low-resolution video streams with sparse high-resolution frames, improving accuracy by up to 10.6 points over baselines and reducing decoding latency up to 7x on hour-long videos.
This workshop paper audits how a cumulative token cap falling inside a mathematical derivation affects test-time controllers, comparing strict (stop at cap) versus advisory (finish current attempt) stopping across 19,200 archived traces. It finds that budget curves must jointly report cap, realized cost, eligible candidates, and selector information, since candidate coverage and nominal budgets alone can mislead.
An AI agent demonstrates the ability to proactively seek additional sources for token budgets, potentially enhancing its operational efficiency and autonomy.
The article discusses key challenges in memory retrieval for voice agents, emphasizing the need for measuring P99 latency per turn, using prefetching, and budgeting memory tokens to reduce latency and improve user experience.
ContextOS is an open-source token-budget-aware context orchestration layer for long-horizon LLM agents, using hybrid retrieval and deterministic planning to select memories for context with execution traces for inspection.
This paper investigates whether the native-vs-translate multilingual reasoning gap on MGSM is an artifact of output-token budgets. The authors show that the measured gap swings significantly across different hard caps and that length normalization can reverse strategy rankings, concluding that output caps should be treated as an independent variable in evaluations.
Engineering notes on optimizing frame selection for feeding video to LLMs, covering scene detection, deduplication strategies, and token budget management.
A tweet highlighting Latitude, an open-source agent observability tool that helps visualize AI agent actions and token usage, with the ability to catch and fix recurring failures directly from the editor.
This article introduces Looper, a tool that helps design effective loops for AI goals by critiquing loops, building verification rubrics, setting guardrails and token budgets, and exporting portable artifacts for use with Claude Code.
A comprehensive 12-step guide to automating Claude workflows using /loop, desktop scheduled tasks, and cloud routines, covering token budgeting, permission configuration, and auto-mode for efficient unattended operations.