Long context fine-tuning of large language models(LLMs) involves training on datasets that are predominantly composed of short sequences and a small proportion of longer sequences. However, existing approaches overlook this long-tail distribution and employ training strategies designed specifically for long sequences. Moreover, these approaches also fail to address the challenges posed by variable sequence lengths during distributed training, such as load imbalance in data parallelism and severe pipeline bubbles in pipeline parallelism. These issues lead to suboptimal training performance and poor GPU resource utilization. To tackle these problems, we propose a chunk-centric training method named ChunkFlow. ChunkFlow reorganizes input sequences into uniformly sized chunks by consolidating short sequences and splitting longer ones. This approach achieves optimal computational efficiency and balance among training inputs. Additionally, ChunkFlow incorporates a state-aware chunk scheduling mechanism to ensure that the peak memory usage during training is primarily determined by the chunk size rather than the maximum sequence length in the dataset. Integrating this scheduling mechanism with existing pipeline scheduling algorithms further enhances the performance of distributed training. Experimental results demonstrate that, compared with Megatron-LM, ChunkFlow can be up to 4.53x faster in the long context fine-tuning of LLMs. Furthermore, we believe that ChunkFlow serves as an effective solution for a broader range of scenarios, such as long context continual pre-training, where datasets contain variable-length sequences.

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a powerful approach to enhancing the reasoning capabilities of Large Language Models (LLMs), yet its underlying mechanisms remain insufficiently understood. In this work, we undertake a pioneering exploration of RLVR through the novel perspective of token entropy patterns, comprehensively analyzing how different tokens influence reasoning performance. By examining token entropy patterns in Chain-of-Thought (CoT) reasoning, we observe that only a small fraction (approximately 20\%) of tokens exhibit high entropy, and these tokens semantically act as critical forks that steer the model toward diverse reasoning pathways. We further demonstrate that moderately increasing the entropy of these high-entropy tokens via decoding temperature adjustments leads to improved performance, quantitatively confirming their role as decision points in reasoning. We ultimately refine RLVR by restricting policy gradient updates to these forking tokens. Despite utilizing only 20\% of tokens, our approach achieves comparable performance to full-gradient updates on the Qwen3-8B base model. Moreover, it demonstrates remarkable improvements on the larger Qwen3-32B base model, boosting AIME'25 scores by 11.04 and AIME'24 scores by 7.71. In contrast, training exclusively on the 80\% lowest-entropy tokens leads to a marked decline in performance. These findings indicate that the efficacy of RLVR primarily arises from optimizing the high-entropy tokens that dictate key reasoning directions. Collectively, our results suggest promising avenues for optimizing RLVR algorithms by strategically leveraging the potential of these high-entropy minority tokens to further enhance the reasoning abilities of LLMs.

Based on the principle of user-centered design, the elderly user needs analysis is an indispensable part of the assistive product design. For this purpose, we designed a questionnaire on assistive products for the elderly and carried out an investigation in China. The questionnaire was aimed at over 60 years of the elder at home. Through the analysis of the questionnaire answers, we tried to understand the physiological function characteristics of the elderly, as well as the assistive product needs, usage intention, usage status, user experience, etc. of the elderly and proposed the application strategies in the design of the assistive products for elderly.

Graph Embedding methods are aimed at mapping each vertex into a low dimensional vector space, which preserves certain structural relationships among the vertices in the original graph. Recently, several works have been proposed to learn embeddings based on sampled paths from the graph, e.g., DeepWalk, Line, Node2Vec. However, their methods only preserve symmetric proximities, which could be insufficient in many applications, even the underlying graph is undirected. Besides, they lack of theoretical analysis of what exactly the relationships they preserve in their embedding space. In this paper, we propose an asymmetric proximity preserving (APP) graph embedding method via random walk with restart, which captures both asymmetric and high-order similarities between node pairs. We give theoretical analysis that our method implicitly preserves the Rooted PageRank score for any two vertices. We conduct extensive experiments on tasks of link prediction and node recommendation on open source datasets, as well as online recommendation services in Alibaba Group, in which the training graph has over 290 million vertices and 18 billion edges, showing our method to be highly scalable and effective.

It is highly desired for existing distributed graph processing systems to support both offline analytics and online queries adaptively. Existing offline graph analytics systems are mostly based on synchronous model. Although achieving high throughput, they suffer relatively high latency in answering simple queries due to synchronization overhead and slow convergence. On the other hand, online graph query systems adopting asynchronous model can response at any time, while incur overwhelmed messages and network packets, making them unable to meet the high throughput demand of offline analytics. In this work, we propose an adaptive asynchronous message processing (AAMP) method, which improves the efficiency of network communication while maintains low latency, to efficiently support offline analytics and online queries in one graph processing framework. We then design GiraphAsync, an implementation of AAMP on top of Apache Giraph, and evaluate it using several representative offline analytics and online queries on large graph datasets. Experimental results show that GiraphAsync gains an up to 10X improvement over synchronous model systems for graph analytics, while performs as well as specialized systems for online graph queries.