AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

AdaSplash-2: Faster Differentiable Sparse Attention

Nuno Gonçalves, Hugo Pitorro, Vlad Niculae, Edoardo Ponti, Lei Li, Andre Martins, Marcos Vinicius Treviso

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Block-sparse attention kernels · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace softmax with alpha-entmax (AdaSplash-2) under NAPELLaMA-3 1B · RULER avg, 32Kaccuracy: 36.8 -> 39.4Improved
replace softmax with alpha-entmax under RoPELLaMA-3 1B · RULER avg, 4K to 32Kaccuracy: 51.4 -> 49.3 (4K); 27.3 -> 23.4 (32K)Hurt
replace softmax with alpha-entmax under NAPELLaMA-3 1B · HELMET ICL avg, 32Kaccuracy: 46.1 -> 62.2Improved
replace softmax with alpha-entmax under RoPELLaMA-3 1B · HELMET ICL avg, 32Kaccuracy: 38.2 -> 39.3Improved
use NAPE instead of RoPE with softmaxLLaMA-3 1B · RULER avg, 32Kaccuracy: 27.3 -> 36.8Improved
replace softmax with alpha-entmax under NAPELLaMA-3 350M · LAMBADAperplexity: 19.23 -> 18.62Improved
replace softmax with alpha-entmax under RoPELLaMA-3 1B · LAMBADAperplexity: 15.01 -> 15.76Hurt
replace softmax with alpha-entmax under NAPELLaMA-3 1B · OLMES 9mcqa avgaccuracy: 53.0 -> 53.1No clear effect
replace softmax with alpha-entmax under RoPELLaMA-3 1B · OLMES 9mcqa avgaccuracy: 52.5 -> 51.2Hurt
initialize tau with histogram (B=4/8/16 bins) in hybrid solver (ablation)AdaSplash-2 solver · Gaussian scores n=4096, alpha=1.5mean absolute error: 10000x closer to tau* after 1 iterationImproved
increase number of histogram bins (ablation)AdaSplash-2 solver · Gaussian scores n=4096, alpha=1.5mean absolute errorImproved
use AdaSplash-2 instead of AdaSplashTriton attention kernel · causal attention, 128K contextstep time: 2.1s -> 1.5sImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy