recurrent-states

Tag

Cards List
#recurrent-states

PACE: Periodic and Channel-Aligned Encoding of Four-Bit Recurrent States

Hugging Face Daily Papers ↗ · 5d ago Cached

PACE 是一种免微调的后训练四比特量化器,用于线性注意力循环状态,通过降低量化频率(每隔几个 token 才舍入一次)并采用列对齐的量化尺度来减少误差,从而在 4B-35B 参数混合模型上以约三分之二的内存达到与八比特 DAMP 相当的性能。

0 favorites 0 likes
#recurrent-states

Expanding Attention: A Taylor-Series Route to Linear-Time Transformers with a Fixed Number of Recurrent States

arXiv cs.LG ↗ · 2026-10-16 Cached

这篇论文提出将注意力的指数核展开为 n 项截断 Taylor 级数,并将每项分解为张量幂内积,从而用固定数量的循环状态实现线性时间 Transformer,实测在长上下文下相比 softmax 注意力有显著加速和常数解码延迟。

0 favorites 0 likes
#recurrent-states

ReaDy: Recurrent States as Dynamic Context for Speculative Drafting in Hybrid Linear Attention Large Language Models

Hugging Face Daily Papers ↗ · 2026-10-14 Cached

ReaDy introduces a speculative decoding method that uses hybrid linear attention recurrent states as dynamic context for drafting, achieving a 3.16x decoding speedup over autoregressive generation with Qwen3.8-27B on an NVIDIA B200 while improving throughput and acceptance length over DSpark across seven benchmarks.

0 favorites 0 likes
← Back to home

Submit Feedback