All pull requests

Pull requests

• perf: store Qwen Image 2.1 prefix cache as F16 with flash attention

  #2044·RChang06 opened Sep 24, 2026·

  ·

• fix: Qwen Image 2.1 alpha output white squares (VAE compute path changed)

  #2043·CarlGao4 opened Sep 24, 2026Contributor·

  ·
  1

  1

• feat: optimize cfg special cases with guidance schdeule

  #2033·stduhpf opened Sep 23, 2026Contributor·

  ·
  2

• feat: add a stand-alone upscale endpoint to the server

  #2026·nbeerbower opened Sep 22, 2026·

  ·
  2

• feat: add native CUDA Sol-Attn support

  #2016·leejet opened Sep 21, 2026Owner·

  ·

• feat: packed int4 convrot model loading and dispatch

  #2010·noctrex opened Sep 20, 2026Contributor·

  ·

• feat: add lanpaint support

  #1978·chaserhkj opened Sep 15, 2026·

  ·
  4

• hexagon: add an experimental Qualcomm FP8 DiT and VAE path

  #1970·happyyzy opened Sep 13, 2026Contributor·

  ·
  5

• chore: Create cuda runtime docker from scratch and drop NCCL

  #1931·fcollonval opened Sep 2, 2026·

  ·

• LTX-2.5 spatial and temporal upscaling only mode

  #1921·jdluzen opened Aug 28, 2026·

  ·

• feat: pool VRAM for streamed layer

  #1906·assouan opened Aug 23, 2026Contributor·

  ·
  2

• Optimize Krea2 Vulkan (target: Strix Halo)

  #1891·pwilkin opened Aug 20, 2026Contributor·

  ·
  4

• Fix ROCm ccache build caching and update to ROCm 10.0.0

  #1890·superm1 opened Aug 20, 2026Contributor·

  ·
  1

• feat(server): report generation progress on async jobs

  #1884·maximedrn opened Aug 16, 2026·

  ·
  3

• feat: support sequential execution for taehv

  #1879·stduhpf opened Aug 13, 2026Contributor·

  ·
  1

• feat: Adds a slim cuda docker image.

  #1867·MicahZoltu opened Aug 10, 2026·

  ·

• Stop MiniMax-H3 aborting on the default cfg-scale and on --vae-on-cpu

  #1862·danielhanchen opened Aug 8, 2026Contributor·

  ·
  2

• refactor: move denoiser code to a separate source file

  #1845·wbruna opened Aug 2, 2026Contributor·

  ·

• Quantization optimization procedures

  #1760·pwilkin opened Jul 7, 2026Contributor·

  ·
  5

• feat: (fix?) try known prefixes to find correspondance for applying imatrix

  #1737·stduhpf opened Jul 3, 2026Contributor·

  ·
  1

• ideogram4: support ref_latents [experimental]

  #1692·stduhpf opened Jun 21, 2026Contributor·

  ·

• feat: add support for init audio allowing Audio to Video (A2V) for LTX2.3

  #1676·LostRuins opened Jun 17, 2026Contributor·

  ·
  1

  1

• feat: accelerate CI build speed by using pre-built pulls.

  #1634·Cyberhan123 opened Jun 11, 2026Contributor·

  ·
  9

• feat: automatically fall back to VAE tiling when an untiled decode exceeds the backend buffer limit

  #1621·RapidMark opened Jun 8, 2026Contributor·

  ·
  20

• towards Nucleus-image support: shared expert only (not working)

  #1445·stduhpf opened Apr 20, 2026Contributor·

  ·