20 papers
From Pareto to Preference: Personalized Test-Time Scaling via Amortized Agentic Policy Discovery
Xinglin Wang, Zishen Liu, Tong Zheng +9
Test-time scaling (TTS) improves the reasoning capabilities of large language models by allocating additional inference computation. Existing approaches to improving TTS efficiency…
Share More, Search Less: Collaborative Parallel Thinking for Efficient Test-Time Scaling
Xinglin Wang, Hao Lin, Shaoxiong Feng +9
Test-Time Scaling (TTS) enhances the reasoning capabilities of large language models by allocating additional inference compute to explore the solution space. However, existing par…
On Time, Within Budget: Constraint-Driven Online Resource Allocation for Agentic Workflows
Xinglin Wang, Zishen Liu, Shaoxiong Feng +9
Agentic systems increasingly solve complex user requests by executing orchestrated workflows, where subtasks are assigned to specialized models or tools and coordinated according t…
Learning More from Less: Unlocking Internal Representations for Benchmark Compression
Yueqi Zhang, Jin Hu, Shaoxiong Feng +9
The prohibitive cost of evaluating Large Language Models (LLMs) necessitates efficient alternatives to full-scale benchmarking. Prevalent approaches address this by identifying a s…
Do Not Waste Your Rollouts: Recycling Search Experience for Efficient Test-Time Scaling
Xinglin Wang, Jiayi Shi, Shaoxiong Feng +8
Test-Time Scaling enhances the reasoning capabilities of Large Language Models by allocating additional inference compute to broaden the exploration of the solution space. However,…
Stitch and Tell: A Structured Multimodal Data Augmentation Method for Spatial Understanding
Hang Yin, Xiaomin He, PeiWen Yuan +5
Existing vision-language models often suffer from spatial hallucinations, i.e., generating incorrect descriptions about the relative positions of objects in an image. We argue that…
Breaking the Self-Confirming Loop: Diagnosing and Mitigating Systemic Reward Bias in Self-Rewarding RL
Chuyi Tan, Peiwen Yuan, Xinglin Wang +8
Reinforcement learning with verifiable rewards (RLVR) efficiently scales the reasoning ability of large language models (LLMs) but is bottlenecked by scarce labeled data. Reinforce…
PatternKV: Flattening KV Representation Expands Quantization Headroom
Ji Zhang, Yiwei Li, Shaoxiong Feng +8
KV cache in autoregressive LLMs eliminates redundant recomputation but has emerged as the dominant memory and bandwidth bottleneck during inference, notably with long contexts and…
Every Rollout Counts: Optimal Resource Allocation for Efficient Test-Time Scaling
Xinglin Wang, Yiwei Li, Shaoxiong Feng +7
Test-Time Scaling (TTS) improves the performance of Large Language Models (LLMs) by using additional inference-time computation to explore multiple reasoning paths through search.…
Mind the Quote: Enabling Quotation-Aware Dialogue in LLMs via Plug-and-Play Modules
Yueqi Zhang, Peiwen Yuan, Shaoxiong Feng +7
Human-AI conversation frequently relies on quoting earlier text-"check it with the formula I just highlighted"-yet today's large language models (LLMs) lack an explicit mechanism f…
Silencer: From Discovery to Mitigation of Self-Bias in LLM-as-Benchmark-Generator
Peiwen Yuan, Yiwei Li, Shaoxiong Feng +7
LLM-as-Benchmark-Generator methods have been widely studied as a supplement to human annotators for scalable evaluation, while the potential biases within this paradigm remain unde…
Speculative Decoding for Multi-Sample Inference
Yiwei Li, Jiayi Shi, Shaoxiong Feng +8
We propose a novel speculative decoding method tailored for multi-sample reasoning scenarios, such as self-consistency and Best-of-N sampling. Our method exploits the intrinsic con…
UniCBE: An Uniformity-driven Comparing Based Evaluation Framework with Unified Multi-Objective Optimization
Peiwen Yuan, Shaoxiong Feng, Yiwei Li +7
Human preference plays a significant role in measuring large language models and guiding them to align with human values. Unfortunately, current comparing-based evaluation (CBE) me…
LLM-Powered Benchmark Factory: Reliable, Generic, and Efficient
Peiwen Yuan, Shaoxiong Feng, Yiwei Li +7
The rapid advancement of large language models (LLMs) has led to a surge in both model supply and application demands. To facilitate effective matching between them, reliable, gene…
Revisiting Self-Consistency from Dynamic Distributional Alignment Perspective on Answer Aggregation
Yiwei Li, Ji Zhang, Shaoxiong Feng +8
Self-consistency improves reasoning by aggregating diverse stochastic samples, yet the dynamics behind its efficacy remain underexplored. We reframe self-consistency as a dynamic d…
Beyond One-Size-Fits-All: Tailored Benchmarks for Efficient Evaluation
Peiwen Yuan, Yueqi Zhang, Shaoxiong Feng +7
Evaluating models on large benchmarks is very resource-intensive, especially during the period of rapid model evolution. Existing efficient evaluation methods estimate the performa…
From Sub-Ability Diagnosis to Human-Aligned Generation: Bridging the Gap for Text Length Control via MARKERGEN
Peiwen Yuan, Chuyi Tan, Shaoxiong Feng +7
Despite the rapid progress of large language models (LLMs), their length-controllable text generation (LCTG) ability remains below expectations, posing a major limitation for pract…
InsBank: Evolving Instruction Subset for Ongoing Alignment
Jiayi Shi, Yiwei Li, Shaoxiong Feng +8
Large language models (LLMs) typically undergo instruction tuning to enhance alignment. Recent studies emphasize that quality and diversity of instruction data are more crucial tha…
Instruction Embedding: Latent Representations of Instructions Towards Task Identification
Yiwei Li, Jiayi Shi, Shaoxiong Feng +6
Instruction data is crucial for improving the capability of Large Language Models (LLMs) to align with human-level performance. Recent research LIMA demonstrates that alignment is…
Tapilot-Crossing: Benchmarking and Evolving LLMs Towards Interactive Data Analysis Agents
Jinyang Li, Nan Huo, Yan Gao +7
Interactive Data Analysis, the collaboration between humans and LLM agents, enables real-time data exploration for informed decision-making. The challenges and costs of collecting…