Search results

Torch 2.10 and python 3.14

Status: Open.

#308 In SHI-Labs/NATTEN;

· karimknaebel opened on Feb 5, 2026

NaN on longer training sequence

Status: Open.

#307 In SHI-Labs/NATTEN;

· coconutruben opened on Feb 5, 2026

temporally causal performance worse than full attention for natten3d

Status: Open.

#306 In SHI-Labs/NATTEN;

· coconutruben opened on Feb 4, 2026

Build issues when multiple torch versions are present

Status: Open.

#304 In SHI-Labs/NATTEN;

· alihassanijr opened on Jan 30, 2026

RoPE vs RPB

Status: Open.

#268 In SHI-Labs/NATTEN;

· Bezdarnost opened on Oct 11, 2025

Implementing NATTEN for 4K image generation with Flux as mentioned in GNA paper

Status: Open.

#264 In SHI-Labs/NATTEN;

· xuanmingShang opened on Sep 18, 2025

Support for Attention Mask in NATTAN (Neighborhood Attention Transformer with Progressive Channel Fusion)

Status: Open.

#263 In SHI-Labs/NATTEN;

· roadroller0501 opened on Sep 5, 2025

reintroduce dropped na2d_qk and na2d_av APIs from 0.17.5

feature-requestFeature requestFeature request

Status: Open.

#262 In SHI-Labs/NATTEN;

· tha-santacruz opened on Aug 29, 2025

NATTEN1D is slower than NATTEN2D?

Status: Open.

#254 In SHI-Labs/NATTEN;

· Yunge6666 opened on Aug 16, 2025

Custom Attention Mask

Status: Open.

#249 In SHI-Labs/NATTEN;

· PaulCouairon opened on Aug 8, 2025

Forward-mode autodiff for fused kernel

feature-requestFeature requestFeature request

Status: Open.

#224 In SHI-Labs/NATTEN;

· Birch-san opened on May 25, 2025

Does NATTEN support double backward?

feature-requestFeature requestFeature request

Status: Open.

#191 In SHI-Labs/NATTEN;

· Luciennnnnnn opened on Jan 4, 2025