LVSA: Training-Free Sparse Attention for Long Video Diffusion

Dense self-attention is the compute and quality bottleneck of long-video diffusion inference: cost grows quadratically with the sequence length, and beyond the training horizon the model converges to near-static output, that is, "frozen" repetitive video. State of the art approaches are either too costly, e.g., they require retraining, or fail to satisfy both performance and quality objectives in a scalable manner. To this end, we introduce Long Video Sparse Attention (LVSA), a training-free model-agnostic block-sparse attention for video diffusion transformers that combines a structured window pattern with rotating global anchors, thus removing the fixed-grid bias which causes long-range temporal artifacts. LVSA, combined with a FlashInfer kernel, reduces compute up to 3.17x on Wan 2.1 1.3B at a 6x horizon, 2.98x on Wan 2.1 14B at a 6x horizon, and 3.33x on HunyuanVideo 1.5 at a 1.5x horizon, compared to dense attention. Beyond reducing compute, LVSA enables HunyuanVideo 1.5 generation at a 2x horizon, which is otherwise out-of-memory on a single GPU. Moreover, LVSA provides speedups up to 2.41x compared to RIFLEx and 3.27x compared to UltraViCo on Wan 2.1 1.3B. To demonstrate applicability across diverse platforms, we apply LVSA on NPUs and achieve speedups up to 2.71x on Wan 2.2 A14B and 3.24x on Wan 2.1 1.3B compared to dense attention. To evaluate quality in a fair way, we introduce VQeval, a tool properly scoring loopy video failures, which instead are rewarded in state of the art evaluators like VBench-Long. LVSA is quality-neutral for generation at training horizon length and quality-positive at extended lengths.

A Constraint Programming Model for Scheduling the Unloading of Trains in Ports: Extended

In this paper, we propose a model to schedule the next 24 hours of operations in a bulk cargo port to unload bulk cargo trains onto stockpiles. It is a problem that includes multiple parts such as splitting long trains into shorter ones and the routing of bulk material through a configurable network of conveyors to the stockpiles. Managing such trains (up to three kilometers long) also requires specialized equipment. The real world nature of the problem specification implies the necessity to manage heterogeneous data. Indeed, when new equipment is added (e.g. dumpers) or a new type of wagon comes in use, older or different equipment will still be in use as well. All these details need to be accounted for. In fact, avoiding a full deadlock of the facility after a new but ineffective schedule is produced. In this paper, we provide a detailed presentation of this real world problem and its associated data. This allows us to propose an effective constraint programming model to solve this problem. We also discuss the model design and the different implementations of the propagators that we used in practice. Finally, we show how this model, coupled with a large neighborhood search, was able to find 24 hour schedules efficiently.