Shijun Yang

Haotian Li, Shijun Yang, Weizhen Qi et al.

Conventional agent systems often struggle in open-ended environments where task distributions continuously drift and external supervision is scarce. Their reliance on static toolsets or offline training lags behind these dynamics, leaving the system's capability boundaries rigid and unknown. To address this, we propose the In-Situ Self-Evolving paradigm. This approach treats sequential task interactions as a continuous stream of experience, enabling the system to distill short-term execution feedback into long-term, reusable capabilities without access to ground-truth labels. Within this framework, we identify tool evolution as the critical pathway for capability expansion, which provides verifiable, binary feedback signals. Within this framework, we develop Yunjue Agent, a system that iteratively synthesizes, optimizes, and reuses tools to navigate emerging challenges. To optimize evolutionary efficiency, we further introduce a Parallel Batch Evolution strategy. Empirical evaluations across five diverse benchmarks under a zero-start setting demonstrate significant performance gains over proprietary baselines. Additionally, complementary warm-start evaluations confirm that the accumulated general knowledge can be seamlessly transferred to novel domains. Finally, we propose a novel metric to monitor evolution convergence, serving as a function analogous to training loss in conventional optimization. We open-source our codebase, system traces, and evolved tools to facilitate future research in resilient, self-evolving intelligence.

Hang Yuan, Bin Yu, Haotian Li et al.

Modern reasoning models, such as OpenAI's o1 and DeepSeek-R1, exhibit impressive problem-solving capabilities but suffer from critical inefficiencies: high inference latency, excessive computational resource consumption, and a tendency toward overthinking -- generating verbose chains of thought (CoT) laden with redundant tokens that contribute minimally to the final answer. To address these issues, we propose Conditional Token Selection (CTS), a token-level compression framework with a flexible and variable compression ratio that identifies and preserves only the most essential tokens in CoT. CTS evaluates each token's contribution to deriving correct answers using conditional importance scoring, then trains models on compressed CoT. Extensive experiments demonstrate that CTS effectively compresses long CoT while maintaining strong reasoning performance. Notably, on the GPQA benchmark, Qwen2.5-14B-Instruct trained with CTS achieves a 9.1% accuracy improvement with 13.2% fewer reasoning tokens (13% training token reduction). Further reducing training tokens by 42% incurs only a marginal 5% accuracy drop while yielding a 75.8% reduction in reasoning tokens, highlighting the prevalence of redundancy in existing CoT.

Shijun Yang, Xiang Zhang, Wanqing Zhao et al.

Prompt learning facilitates the efficient adaptation of Vision-Language Models (VLMs) to various downstream tasks. However, it faces two significant challenges: (1) inadequate modeling of class embedding distributions for unseen instances, leading to suboptimal generalization on novel classes; (2) prevailing methodologies predominantly confine cross-modal alignment to the final output layer of vision and text encoders, which fundamentally limits their capacity to preserve topological consistency with pre-trained multi-modal embedding spaces. To this end, we introduce MuGCP (Multi-modal Mutual-Guidance Conditional Prompt Learning), a novel paradigm designed for conditional prompt generation. MuGCP leverages Multi-modal Large Language Models (MLLMs) as conditional prompt learners to adaptively generate Semantic Conditional Prompts (SCP) that incorporate rich, fine-grained high-level semantic knowledge for image instances. To ensure effective alignment and interaction across the multi-modal space of Vision-Language Models (VLMs), we introduce the Attention Mutual-Guidance (AMG) module, which facilitates interactions between visual and semantic information. Through mutual guidance, the AMG module generates Visual Conditional Prompts (VCP), enhancing the model's performance in multi-modal tasks. Additionally, we present a Multi-Prompt Fusion (MPF) mechanism that integrates SCP and VCP with contextual prompts, ensuring seamless coordination among the different prompts and enhancing the modeling of class embeddings and instance-specific knowledge. Our MuGCP outperforms existing state-of-the-art methods on 14 different datasets. The code will be made available after publication.