7 papers
CoTrace: Data Recipes for Training Terminal Agents with Harness-Model Co-Evolution
Jixuan Chen, Jiaxin Zhang, Qinyuan Ye +11
Terminal-agent capability depends jointly on model weights and the runtime harness that formats prompts, binds tools, and handles error recovery. Existing harness-model co-evolutio…
DeliveryGym: An RL Environment for Long-Horizon Embodied Agent Planning with Adaptive Curriculum
Haoqiang Kang, Yiming Zhang, Yiyang Guo +5
Executable environments enable LLM agents to learn from the consequences of their actions. For embodied agents, those consequences extend beyond whether the current task succeeds:…
Auto-Configuring Scientific Simulators with Lightweight Coding-Agent Adapters
Matthew Ho, Brian Liu, Jixuan Chen +2
Configuring an advanced scientific simulator, translating a modeling goal into a valid, runnable input deck, is a persistent bottleneck that costs domain scientists hours to days.…
AgentSpec: Understanding Embodied Agent Scaffolds Through Controlled Composition
Jixuan Chen, Jianzhi Shen, Haoqiang Kang +10
LLM agents are increasingly built not as single model calls, but as scaffolded systems that combine reasoning, memory, reflection, action execution, and learning. While such scaffo…
Latent Reasoning with Normalizing Flows
Guancheng Tu, Xiangjun Fu, Suhao Yu +5
Large language models often improve reasoning by generating explicit chain-of-thought (CoT), demonstrating the importance of intermediate computation. However, textual CoT forces t…
MOSAIC: More LLMs, Less Cost through Collaborative Reasoning for Compiler Optimizations
Annabelle Sujun Tang, Christopher Priebe, Lianhui Qin +1
LLM-guided compiler optimization has recently shown promise. However, current LLM-guided optimizers query a single LLM at every search step, so their cost grows with both that mode…
REASONING COMPILER: LLM-Guided Optimizations for Efficient Model Serving
Annabelle Sujun Tang, Christopher Priebe, Rohan Mahapatra +2
While model serving has unlocked unprecedented capabilities, the high cost of serving large-scale models continues to be a significant barrier to widespread accessibility and rapid…