Junhao Ruan

HybridThinker: Efficient Chain-of-Thought Reasoning via Compressed Memory and Transient Thought Steps

Xin Liu (43569), Runsong Zhao, Xinyu Liu, Junhao Ruan et autres

Extended chain-of-thought (CoT) traces improve LLM reasoning but incur substantial computational and memory costs. While existing CoT compression methods mitigate this by condensing thought steps into compact representations via memory tokens and retaining only these representations at inference time, the loss of …

SubQRAG: Sub-Question Driven Dynamic Graph Rag

Jiaoyang Li, Junhao Ruan, S. Q. Tang, Yuan Ge et autres

Graph Retrieval-Augmented Generation (Graph RAG) is effective at building a knowledge graph (KG) to connect disparate facts across a wide document corpus. However, this broad-view approach often lacks the deep structured reasoning required for complex multi-hop question answering (QA), leading to incomplete …

cn (code pays fourni par la source)

Causal Autoregressive Diffusion Language Model

Junhao Ruan, Bei Li, Yongjing Yin, Pengcheng Huang et autres

In this work, we propose Causal Autoregressive Diffusion (CARD), a novel framework that unifies the training efficiency of ARMs with the high-throughput inference of diffusion models. CARD reformulates the diffusion process within a strictly causal attention mask, enabling dense, per-token supervision in …