Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) exhibit fundamentally different behaviors in enhancing multi-task reasoning for large language models (LLMs). Our preliminary experiments revealed a phenomenon: SFT suffers from severe task conflicts under multi-stage training, whereas RL enables stable coexi...
Kejian Zhu, Zhuo-Ran Jin, Shangqing Tu et al.· 0 citations
This work introduces AREX, a family of Recursively Self-Improving (RSI) deep research agents that substantially outperforms comparable-scale baselines and remains competitive with models using substantially more activated parameters.
Shuqi Lu, Chaofan Li, Kun Luo et al.· arXiv.org· 2 citations· ⚡1
SwarmBench is proposed, a benchmark that evaluates model performance from multiple perspectives, including accuracy, efficiency, cost, and process quality, and SwarmExp is proposed, a simple yet effective method based on experience extraction and experience replay, which consistently improves the orchestration performa...
Jin Gao, Zhuoran Jin, Tianyi Men et al.· 0 citations
By prioritizing operator over aggressive operand precision, the ATOM-synthesized data outperforms rigorous baselines (e.g., +3.1% gain over LIMA), suggesting that operator diversity matters more than operand precision.
Jia-Xiang Liu, Chenhao Yuan, Shunnan Xu et al.· 0 citations
RuleWeaver is introduced, a benchmark construction framework for evaluating rule-centered scenario reasoning that starts from corpus-derived IF-THEN Meta Rules, progressively augments them into complex rules, and composes these rules into rule-centered scenario QA instances.
Bo-Han Yu, Shi-Yang Li, Pengfei Cao et al.· 0 citations
DynaRule is proposed, an end-to-end framework that injects the given rules into the KV cache and turns retrieval into an internal, learnable, step-wise process, and can re-attend to the most relevant rules at each step, dynamically replacing outdated ones to support more stable multi-step reasoning.
Bo-Han Yu, Pengfei Cao, Chen Han et al.· 1 citation
This work systematically study weight-only post-training quantization across bit-widths, quantization methods, model scales and downstream tasks and establishes that quantization degradation is governed by how errors are introduced at the source and how they accumulate across the network.
Chen-Xi Zhou, Pengfei Cao, Jin Ye et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
