Issues

• Community integration: ComfyUI-SparkDiffusion + RTX 5090 / SM120 reproduction results

  AlibabaResearch/SparkDiffusion#15 · Filter by author hiroki-abe-58 opened on Oct 1, 2026
  3

• Choosing a sufficient sparse warm-up budget for a new setting

  AlibabaResearch/SparkDiffusion#14 · Filter by author HuangZhengyang-2004 opened on Sep 30, 2026
  1

• How should inference-time sparsity relate to the checkpoint's training setting?

  AlibabaResearch/SparkDiffusion#13 · Filter by author HuangZhengyang-2004 opened on Sep 30, 2026
  1

• Appreciation: Thank you for open-sourcing SparkDiffusion!

  AlibabaResearch/SparkDiffusion#12 · Filter by author HuangZhengyang-2004 opened on Sep 29, 2026

• Small documentation consistency notes while reading the agent skill docs (naming alignment with the README)

  AlibabaResearch/SparkDiffusion#11 · Filter by author kunkunblueberry opened on Sep 29, 2026
  1

• The pieces that make a 200x+ result reproducible by people outside the team

  AlibabaResearch/SparkDiffusion#10 · Filter by author kunkunblueberry opened on Sep 29, 2026

• Appreciation: engineering design of the sparse-attention plugin path, the fused operator suite, and the checkpoint-loading contract

  AlibabaResearch/SparkDiffusion#9 · Filter by author kunkunblueberry opened on Sep 29, 2026

• 这个有适配mxfp8的sparse attention算子呢？或者有没有在mxfp8上测试过

  AlibabaResearch/SparkDiffusion#8 · Filter by author ultism opened on Sep 29, 2026
  1

• Reproduction report: 1.3B FP8 RoLA inference on RTX 5080 (SM120)

  AlibabaResearch/SparkDiffusion#7 · Filter by author AndreiB180 opened on Sep 28, 2026
  1