llm-kb
← Back to research
Paper

KV-PRM: Efficient Process Reward Modeling via KV-Cache Transfer for Multi-Agent Test-Time Scaling

A new paper introduces KV-PRM, a process reward model that reuses KV-cache from the base model to avoid re-encoding entire trajectories, reducing scoring cost from quadratic to linear in sequence length. This enables efficient test-time scaling for long-context multi-agent systems.

0 engagement·1 source·Fri, Jul 10, 2026, 07:16 AM
The paper 'KV-PRM: Efficient Process Reward Modeling via KV-Cache Transfer for Multi-Agent Test-Time Scaling' (arXiv, 2026-07-10) addresses the computational bottleneck of existing text-based Process Reward Models (PRMs) that re-encode the entire trajectory from scratch, leading to O(L²) cost. KV-PRM transfers KV-cache from the base model's forward pass, eliminating redundant computation and achieving O(L) scoring cost. This makes PRMs practical for long-context multi-agent rollouts, significantly boosting LLM-based multi-agent capabilities via test-time scaling.

Entities

KV-PRM(model)Process Reward Model(concept)Test-Time Scaling(concept)KV-Cache Transfer(concept)

Related

PaperTue, Jul 7, 2026, 05:26 PM

Two papers propose token-adaptive KV cache compression for long-context LLMs

Two arXiv papers from July 7, 2026 introduce token-adaptive KV cache compression methods for long-context LLM inference. DepthWeave-KV factorizes key/value states across neighboring layers using shared low-rank bases with token-specific residuals. FreqDepthKV uses shared low-frequency depth components and sparse high-frequency residuals, with an online probe assigning attention heads to different cache modes. Both aim to reduce memory bandwidth while preserving retrieval and reasoning quality.

0 engagement·2 sources·arxiv
arXiv
PaperThu, Jul 9, 2026, 05:26 PM

Researchers propose memory agent to combat behavioral state decay in long-horizon tasks

A new arXiv paper identifies 'behavioral state decay' as a failure mode where decision-relevant information gets buried in long trajectories. The authors propose a separate memory agent that actively updates a structured memory bank alongside an unmodified action agent, rather than relying on passive retrieval.

0 engagement·1 source·arxiv
arXiv
PaperFri, Jul 10, 2026, 12:42 AM

Paper proposes correlation-aware bandits with surrogate rewards for LLM routing

A new arXiv paper introduces contextual bandit algorithms that leverage surrogate reward signals from machine learning models to improve LLM routing decisions. The approach accounts for inter-arm correlations and noisy auxiliary rewards, addressing limitations of classical bandits that assume conditional independence.

0 engagement·1 source·arxiv
arXiv
BenchmarkThu, Jul 9, 2026, 09:56 PM

Long-Horizon-Terminal-Bench: New Benchmark Tests AI Agents on Long-Horizon Tasks with Dense Rewards

Researchers introduced Long-Horizon-Terminal-Bench, a benchmark of 46 long-horizon tasks across nine categories, including experiment reproduction and software engineering. Unlike existing benchmarks that evaluate only final outcomes, it uses dense reward signals to measure intermediate progress, providing a more complete picture of agent capability.

0 engagement·1 source·arxiv
arXiv
PaperFri, Jul 10, 2026, 08:48 AM

Paper proposes modeling forgetting as interference between tasks in continual learning

A new arXiv paper argues that forgetting in continual learning should be modeled directly as interference between tasks, rather than relying on post-hoc mechanisms like replay or regularization. The authors show that in the frozen-feature regime, forgetting equals the interference energy induced on the old task, and they recover this quantity in deep networks via path-averaged curvature with minimal extra computation.

0 engagement·1 source·arxiv
arXiv