Skip to content

Pull requests: hao-ai-lab/FastVideo

Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort

Pull requests list

[perf] FastH3 Preview inference on a PCIe-attached B200 node: 15.2 s to 6.3 s per generation at 4 GPUs, from the frozen VAEs' offload copies scope: inference Inference pipeline, serving, CLI scope: model Model architecture (DiTs, encoders, VAEs) type: perf Performance improvement
#1867 opened Sep 17, 2026 by TarzanZhao Loading…
3 of 6 tasks
[kernel] sm_100a CUDA backward for VSA block-sparse attention, 128-token blocks scope: kernel CUDA kernels, fastvideo-kernel
#1866 opened Sep 17, 2026 by lpc0220 Contributor Loading…
4 tasks done
[feat] Import MiniMax H3 Comfy NVFP4-AWQ text encoder scope: infra CI, tests, Docker, build scope: model Model architecture (DiTs, encoders, VAEs) type: feat New feature or capability
#1865 opened Sep 17, 2026 by LeslieWylie Contributor Loading…
[feat] Add opt-in Atlas Cloud prompt provider for Dreamverse type: feat New feature or capability
#1864 opened Sep 17, 2026 by binyangzhu000-sudo Loading…
3 of 4 tasks
[bugfix] Enable SP for LongCat BSA refinement scope: inference Inference pipeline, serving, CLI scope: model Model architecture (DiTs, encoders, VAEs) type: bugfix Bug fix
#1862 opened Sep 16, 2026 by Livinfly Loading…
6 tasks done
[perf] FP8: fuse the dynamic activation quantization into Triton kernels scope: docs Documentation scope: infra CI, tests, Docker, build scope: model Model architecture (DiTs, encoders, VAEs) type: perf Performance improvement
#1861 opened Sep 16, 2026 by stayinalive181 Contributor Loading…
[kernel] VSA block-sparse attention: HIP autotune configs for gfx950 (1.22-1.35x on MI355X) scope: kernel CUDA kernels, fastvideo-kernel
#1860 opened Sep 16, 2026 by stayinalive181 Contributor Loading…
[bugfix] Keep the VSA-H3 tile buffer out of the autograd graph scope: attention Attention backends (VSA, STA, Flash, etc.) scope: infra CI, tests, Docker, build type: bugfix Bug fix
#1858 opened Sep 16, 2026 by Jerry-XY Loading…
4 of 6 tasks
[bugfix] Preserve video frames when decoded audio is shorter scope: inference Inference pipeline, serving, CLI scope: infra CI, tests, Docker, build type: bugfix Bug fix
#1857 opened Sep 16, 2026 by LeslieWylie Contributor Loading…
[docs] Define Dreamverse system prompts for T2VA, FL2VA and Ref2VA scope: docs Documentation type: docs Documentation only
#1855 opened Sep 15, 2026 by jayzou3773 Draft
4 of 5 tasks
[feat] ROCm: route the video sparse attention backends to their Triton kernels scope: infra CI, tests, Docker, build type: feat New feature or capability
#1851 opened Sep 15, 2026 by stayinalive181 Contributor Loading…
5 of 6 tasks
[bugfix] ROCm: detect the platform from a HIP torch build when amdsmi is unavailable scope: infra CI, tests, Docker, build type: bugfix Bug fix
#1849 opened Sep 15, 2026 by stayinalive181 Contributor Loading…
4 tasks done
[bugfix]: isolate cancelled streaming requests from GPU result readers scope: inference Inference pipeline, serving, CLI scope: infra CI, tests, Docker, build type: bugfix Bug fix
#1848 opened Sep 15, 2026 by Gxj230958 Loading…
[perf] FastH3 Preview inference on a PCIe-attached B200 node: 15.0 s to 3.9 s per generation at 4 GPUs, mostly from the VAE offload path needs-rebase PR has merge conflicts scope: docs Documentation scope: inference Inference pipeline, serving, CLI scope: infra CI, tests, Docker, build scope: model Model architecture (DiTs, encoders, VAEs) type: perf Performance improvement
#1836 opened Sep 9, 2026 by TarzanZhao Draft
4 of 6 tasks
[feat] Add Dreamverse multimodal inputs and H3 routing scope: docs Documentation type: feat New feature or capability
#1835 opened Sep 9, 2026 by jayzou3773 Draft
4 tasks
[feat] Add FlashInfer attention backend support for MiniMax H3 scope: attention Attention backends (VSA, STA, Flash, etc.) scope: docs Documentation scope: infra CI, tests, Docker, build scope: model Model architecture (DiTs, encoders, VAEs) type: feat New feature or capability
#1832 opened Sep 8, 2026 by klhhhhh Contributor Loading…
[feat] Add MiniMax-H3 disaggregated inference across two DGX Sparks scope: docs Documentation scope: inference Inference pipeline, serving, CLI scope: infra CI, tests, Docker, build type: feat New feature or capability
#1830 opened Sep 8, 2026 by Satyam-53 Collaborator Loading…
3 of 6 tasks
[feat] MiniMax H3: block-FP8 text encoder on Hopper, its converter, and a 4xH100 FastH3 profile scope: inference Inference pipeline, serving, CLI scope: infra CI, tests, Docker, build scope: model Model architecture (DiTs, encoders, VAEs) type: feat New feature or capability
#1829 opened Sep 7, 2026 by stayinalive181 Contributor Loading…
5 of 6 tasks
[perf] MiniMax H3: return uint8 frames from the decode worker scope: inference Inference pipeline, serving, CLI scope: infra CI, tests, Docker, build scope: training Training pipeline, methods, configs type: perf Performance improvement
#1828 opened Sep 7, 2026 by stayinalive181 Contributor Loading…
5 of 6 tasks
[perf]FlashInfer cuDNN batched prefill attention scope: attention Attention backends (VSA, STA, Flash, etc.) scope: docs Documentation scope: infra CI, tests, Docker, build type: perf Performance improvement
#1827 opened Sep 7, 2026 by klhhhhh Contributor Loading…
[feat] Add phase-aware modular training performance metrics scope: docs Documentation scope: infra CI, tests, Docker, build scope: training Training pipeline, methods, configs type: feat New feature or capability
#1826 opened Sep 7, 2026 by Jack47 Loading…
ProTip! Updated in the last three days: updated:>2026-09-16.