Search results

Question on planning-stage optimization for VariableBlockSparseAttentionWrapper

Status: Open.

#68 In svg-project/Sparse-VideoGen;

· tie-pilot-qxw opened on Jan 16, 2026

Triton/CUDA Illegal Memory Access in triton_rmsnorm_forward with Wan2.1-T2V-1.3B-Diffusers

Status: Open.

#55 In svg-project/Sparse-VideoGen;

· 1717Li opened on Sep 29, 2025

Diffsynth Studio Support

Status: Open.

#43 In svg-project/Sparse-VideoGen;

· jxjx12138 opened on Jul 11, 2025

A potentional issue on Blackwell GPUs (B200) for the placement operator

Status: Open.

#40 In svg-project/Sparse-VideoGen;

· wenhao728 opened on Jun 18, 2025

No obvious improvement observed on A100.

Status: Open.

#38 In svg-project/Sparse-VideoGen;

· Chuge0335 opened on May 27, 2025

No efficiency gains on A100

Status: Open.

#36 In svg-project/Sparse-VideoGen;

· jhl13 opened on May 23, 2025

The orig_wan implementation has a different block mask

Status: Open.

#29 In svg-project/Sparse-VideoGen;

· m-richa opened on May 6, 2025

Compatibility with other acceleration methods, such as TeaCache, Comfy-Wavespeed, torch.compile, and/or attention optimization

Status: Open.

#12 In svg-project/Sparse-VideoGen;

· Kazuki-Yoda opened on Apr 4, 2025

Does SVG Support Sequence Parallelism

Status: Open.

#9 In svg-project/Sparse-VideoGen;

· sywang0111 opened on Mar 25, 2025

Can this be converted into a Comfy node?

Status: Open.

#3 In svg-project/Sparse-VideoGen;

· CustodialMatter opened on Mar 13, 2025