Qiaolin Yu

Pin Qian, Su Wang, Xiaoyuan Wang et al.

Cited RAG evaluation often treats visible sources as a grounding signal, but a real, topically relevant citation can still under-warrant the attached wording. We study this diagnostic failure as citation laundering: a related source is presented as warrant for an over-strong claim. We introduce FORCEBENCH, a contrastive stress test for evidence-force calibration. Each item holds a cited passage fixed and pairs an evidence-calibrated claim with a localized force-raised variant across five operational axes: relation, modality, scope, temporal validity, and numeric specificity. A calibrated evaluator should score the evidence-calibrated claim higher. Headline experiments use a fixed, locality-filtered 198-pair evaluation set. A citation-presence sanity check is uninformative by design; token and entity overlap still violate monotonicity on 32.8--36.4% of pairs. Across four reported model judges, standard generic support prompting is insufficient for this force-calibration stress test (aggregate MVR 47.2%), while explicit warrant-strength prompting lowers MVR to 24.5% but remains imperfect. We release the benchmark, prompts, outputs, and plug-in pipeline so citation evaluators can report monotonicity violation rate and force sensitivity alongside conventional support metrics.

Yuzhen Zhou, Jiajun Li, Yusheng Su et al.

Reinforcement learning (RL) has become a cornerstone in advancing large-scale pre-trained language models (LLMs). Successive generations, including GPT-o series, DeepSeek-R1, Kimi-K1.5, Grok 4, and GLM-4.5, have relied on large-scale RL training to enhance reasoning and coding capabilities. To meet the community's growing RL needs, numerous RL frameworks have been proposed. However, RL training remains computationally expensive, with rollout generation accounting for more than 90% of total runtime. In addition, its efficiency is often constrained by the long-tail distribution of rollout response lengths, where a few lengthy responses stall entire batches, leaving GPUs idle and underutilized. As model and rollout sizes continue to grow, this bottleneck increasingly limits scalability. To address this challenge, we propose Active Partial Rollouts in Reinforcement Learning (APRIL), which mitigates long-tail inefficiency. In the rollout phase, APRIL over-provisions rollout requests, terminates once the target number of responses is reached, and recycles incomplete responses for continuation in future steps. This strategy ensures that no rollouts are discarded while substantially reducing GPU idle time. Experiments show that APRIL improves rollout throughput by 22.5% on average (at most 44%) across commonly used RL algorithms (GRPO, DAPO, GSPO), accelerates convergence, and achieves 2.1% on average(at most 8%) higher final accuracy across tasks. Moreover, APRIL is both framework and hardware agnostic, already integrated into the slime RL framework, and deployable on NVIDIA and AMD GPUs alike. Taken together, this work unifies system-level and algorithmic considerations in proposing APRIL, with the aim of advancing RL training efficiency and inspiring further optimizations in RL systems. Our codebase is available at https://github.com/RLsys-Foundation/APRIL

Ji Liu, Beichen Ma, Qiaolin Yu et al.

Federated Learning (FL) is a promising distributed machine learning approach that enables collaborative training of a global model using multiple edge devices. The data distributed among the edge devices is highly heterogeneous. Thus, FL faces the challenge of data distribution and heterogeneity, where non-Independent and Identically Distributed (non-IID) data across edge devices may yield in significant accuracy drop. Furthermore, the limited computation and communication capabilities of edge devices increase the likelihood of stragglers, thus leading to slow model convergence. In this paper, we propose the FedDHAD FL framework, which comes with two novel methods: Dynamic Heterogeneous model aggregation (FedDH) and Adaptive Dropout (FedAD). FedDH dynamically adjusts the weights of each local model within the model aggregation process based on the non-IID degree of heterogeneous data to deal with the statistical data heterogeneity. FedAD performs neuron-adaptive operations in response to heterogeneous devices to improve accuracy while achieving superb efficiency. The combination of these two methods makes FedDHAD significantly outperform state-of-the-art solutions in terms of accuracy (up to 6.7% higher), efficiency (up to 2.02 times faster), and computation cost (up to 15.0% smaller).