7 papers · 1 filter
AdaMTP: An Adaptive Training Paradigm for Multi-Token Prediction
Ziqiang Cui, Han Shi, Bowei He +8
Multi-Token Prediction (MTP) has emerged as an effective paradigm that augments a shared Large Language Model backbone with auxiliary heads, training the model to predict several f…
Reason Only When Needed: Efficient Generative Reward Modeling via Model-Internal Uncertainty
Chao Xue, Yao Wang, Mengqiao Liu +11
Recent advancements in the Generative Reward Model (GRM) have demonstrated its potential to enhance the reasoning abilities of LLMs through Chain-of-Thought (CoT) prompting. Despit…
Beyond Semantic Relevance: Counterfactual Risk Minimization for Robust Retrieval-Augmented Generation
Peiyang Liu, Qiang Yan, Ziqiang Cui +3
Standard Retrieval-Augmented Generation (RAG) systems predominantly rely on semantic relevance as a proxy for utility. However, this assumption collapses in realistic decision-maki…
Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Models
Chao Xue, Yao Wang, Mengqiao Liu +11
Supervised Fine-Tuning (SFT) is the standard approach for adapting large language models (LLMs) to downstream tasks. However, we observe a persistent failure mode: even after conve…
StructKV: Preserving the Structural Skeleton for Scalable Long-Context Inference
Zhirui Chen, Peiyang Liu, Ling Shao
As Large Language Models (LLMs) scale to support context windows exceeding one million tokens, the linear growth of Key-Value (KV) cache imposes severe memory capacity and bandwidt…
ToolSafe: Enhancing Tool Invocation Safety of LLM-based agents via Proactive Step-level Guardrail and Feedback
Yutao Mou, Zhangchi Xue, Lijun Li +4
While LLM-based agents can interact with environments via invoking external tools, their expanded capabilities also amplify security risks. Monitoring step-level tool invocation be…
Less Is More: Elevating RAG via Performance-Driven Context Compression
Ziqiang Cui, Yunpeng Weng, Xing Tang +7
Retrieval-Augmented Generation (RAG) has emerged as a promising paradigm for improving the timeliness of knowledge updates and the factual accuracy of large language models. Howeve…