3 papers
cs.LG2026
Blurry Window Attention
Axel Laborieux, Christos Sourmpis, Juan Gabriel Kostelec +1
The Softmax Attention operation in Transformer language models has a quadratic complexity in the sequence length and a growing state size in the form of KV cache, which becomes a b…
cs.CL2026
When Perplexity Lies: Generation-Focused Distillation of Hybrid Sequence Models
Juan Gabriel Kostelec, Xiang Wang, Axel Laborieux +2
Converting a pretrained Transformer into a more efficient hybrid model through distillation offers a promising approach to reducing inference costs. However, achieving high-quality…
cs.AI2026
AllMem: A Memory-centric Recipe for Efficient Long-context Modeling
Ziming Wang, Xiang Wang, Kailong Peng +5
Large Language Models (LLMs) encounter significant performance bottlenecks in long-sequence tasks due to the computational complexity and memory overhead inherent in the self-atten…