arXiv:2611.06828v1 Announce Type: new Abstract: Gated-delta linear attention layers such as Gated DeltaNet (GDN), Kimi Delta Attention (KDA), and GDN-2 are tuned for GPUs, where small tensor-core til...
arXiv:2606. 13392v1 Announce Type: new Abstract: Ultra-long-context capability is becoming indispensable for frontier LLMs: agentic workflows, repository-scale code reasoning, and persistent memory all require the model to jointly attend over hundreds of thousands to millions of tokens, yet the quadratic cost of softmax attention makes this untenable at deployment scale.
Muir is a new linear‑attention architecture that consolidates gating mechanisms into a low‑rank projection and uses a matched‑filter readout to re‑weight state rows. It improves over GDN‑2 and Transformers across a wide range of model sizes, lowering perplexity, boosting commonsense accuracy, and enhancing extractive recall while maintaining comparable throughput. The design also includes a numerically stable parallel update for efficient training on large chunk sizes.
arXiv:2606. 20005v1 Announce Type: cross Abstract: Attention distillation, which trains one attention distribution to match another by minimizing their Kullback-Leibler (KL) divergence, is widely used in knowledge distillation, model compression, continual learning, and sparse-attention LLM training.
arXiv:2608.30386v1 Announce Type: cross Abstract: Hybrid linear-attention architectures have recently scaled to large open-weight models, offering quality competitive with full attention while substa...
arXiv:2607. 14568v1 Announce Type: cross Abstract: A companion study ran a 35B mixture-of-experts model on a 2011 NVIDIA Tesla C2075 (Fermi, sm_20, 6GB) as a GPU-prefill/CPU-decode hybrid, because the 4-bit model did not fit in device memory (arXiv:2606.
FlyWheel introduces optimized TPU kernels for softmax and linear attention, reducing vector work and improving matrix operand reuse. By overlapping independent computations, FlyWheel achieves up to 4.14× speedups over the RPA kernel and 1.80× over Splash Attention, reaching 81% of peak BF16 throughput on TPU v6e. It also boosts Gated DeltaNet prefill by up to 3.61× and enhances vLLM serving of Qwen3.8-27B prefill throughput by up to 1.56× while preserving accuracy.
The paper introduces Group-Query Latent Attention (GQLA), a variant of Multi-head Latent Attention that shares group‑sharded up‑projections to enable a joint latent representation and a shardable expanded cache. GQLA supports both MQA‑absorb and GQA execution using the same weights and existing kernels, and roofline analysis shows how absorption trades traffic for computation while GQA shards KV groups. Experiments on DeepSeek‑V3.1‑Base and other models demonstrate that GQLA achieves comparable training losses to MLA, maintains high accuracy after checkpoint conversion, and improves output throughput by up to 35% on eight H20‑3e GPUs, while also allowing larger batch sizes due to a smaller per‑rank cache.
arXiv:2609.13612v1 Announce Type: new Abstract: Modern AI systems are built on the Transformer architecture, whose core operation, attention, accounts for the majority of computation and memory cost....
DeltaLLM is a training‑free framework that accelerates on‑device large language model inference by introducing delta‑key sparse attention and a vectorized structured sparse online softmax. It reuses delta keys across query heads, preserves exact attention near the diagonal while approximating non‑local regions, and exploits hardware‑friendly sparsity with a specialized vector‑processor architecture. Experiments on BitNet‑b1.58‑2B‑4T and LLaMA3.2‑1B‑Instruct show up to 60% delta‑key sparsity, 63% structured attention‑score sparsity, and a 1.62× speedup on the KV260 FPGA with minimal accuracy loss.
The paper investigates how block‑diffusion language models can use a constant‑size cache to enable efficient parallel decoding. By employing sequence mixers that summarize completed blocks into a reusable state and a block‑causal training objective, the authors pretrain three 3B block‑diffusion denoisers (attention, Mamba, and hybrid) on 300 B tokens. The resulting state‑space cache remains O(1) in memory and latency regardless of context length, yielding significant speed‑up and memory savings compared to traditional attention‑based caches, especially at very long sequences.
arXiv:2606. 04511v1 Announce Type: cross Abstract: Sparse attention reduces compute and memory bandwidth for long-context LLM inference.