Tag
CBPO introduces a contrastive branch policy optimization method for fine-grained credit assignment in reinforcement learning with verifiable rewards, enhancing language model performance in tool-integrated reasoning tasks across multiple benchmarks.
Proposes HiDiffTIR, a hierarchical difficulty-aware policy optimization framework for multi-turn tool-integrated reasoning in LLM agents, improving performance and tool invocation accuracy through fine-grained credit assignment.
TurnSight introduces a turn-level hindsight self-distillation framework for tool-integrated reasoning, providing dense supervision via execution-conditioned hindsight and adaptive RL advantage modulation.
ToolVerse is a framework that automatically builds massive executable agent training environments from 422 real-world MCP environments containing 4438 tools, and proposes a task design strategy using Dynamic Unlocking Sampling Algorithm to generate long-horizon tasks, along with a Turn-Aware Relative Advantage algorithm for credit assignment in agentic reinforcement learning.
The ToolMaze benchmark evaluates LLM agents' ability to handle real-world tool failures, revealing that implicit semantic failures cause the largest performance drops and that dynamic replanning remains a critical bottleneck not addressed by scaling or prompting.
This paper introduces ThinC (Thinking in Code), a framework where language models use code blocks exclusively for reasoning after a brief natural language planning step, outperforming existing tool-integrated reasoning baselines on math benchmarks.