Papers arxiv:2605.01214 Agentic AI Systems Should Be Designed as Marginal Token Allocators Published on May 2 Submitted by siqi zhu on May 5 University of Illinois at Urbana-Champaign Authors: Siqi Zhu Abstract Agentic AI systems should be evaluated as marginal token allocation economies rather tha… …
… These results suggest that much of the uncertainty information captured by multi-sample agreement is already available in the model's initial token distribution. We argue that phi first should be reported as a default low-cost baseline before invoking sampling-based uncertainty estimation . …
… AI-generated summary LLM inference is still evaluated mainly as a model or software problem: accuracy, latency, throughput, and hardware utilization. …
Papers arxiv:2605.07210 DiffRetriever: Parallel Representative Tokens for Retrieval with Diffusion Language Models Published on May 8 Submitted by shuai wang on May 11 The Information Engineering Lab Authors: , , Shengyao Zhuang , , Abstract DiffRetriever enables efficient multi-token retrieval usi… …
Papers arxiv:2605.06105 Shallow Prefill, Deep Decoding: Efficient Long-Context Inference via Layer-Asymmetric KV Visibility Published on May 7 Submitted by jeongseokoh on May 11 Authors: Jungsuk Oh , , , , Abstract SPEED is a phase-asymmetric KV-visibility policy that reduces long-context inference… …
… On GSM8K at a budget of 200 tokens, LenVM maintains 63% accuracy compared to 6 percent for token budget baseline. …
… AI-generated summary DeepSeek Sparse Attention DSA sets the state of the art for fine-grained inference-time sparse attention by introducing a learned token-wise indexer that scores every prefix token and selects the most relevant ones for the main attention. …
… We further provide theoretical analysis showing that preferentially retaining useful tokens reduces attention dilution , and we justify geometric retention as a query-agnostic proxy for future utility. …
… Published on May 9 Submitted by Tianyu Yu on May 12 Authors: Kechen Fang , , Chongyi Wang , , Tianyu Yu , Abstract Efficient visual encoding for high-resolution inputs in multimodal large language models is achieved through slice-based encoding and intra-ViT early compression, reducing computationa… …
… While diffusion language models attempt to break this barrier via parallel generation, they suffer from significant performance degradation, high training costs, and a lack of rigorous convergence guarantees. …