Commits

Commit history

1. Commits on Oct 6, 2026

   • chore: resolve MSVC warnings in ggml extensions

     leejet

     committed
     a1ded76View commit details

     Copy full SHA for a1ded76

     Browse repository at this point

   • sync: update frontend

     leejet

     committed
     8cc5ad8View commit details

     Copy full SHA for 8cc5ad8

     Browse repository at this point

   • docs: document video reference image support and constraints

     leejet

     committed
     4a89229View commit details

     Copy full SHA for 4a89229

     Browse repository at this point

   • refactor: escape newlines in prompt logs and clarify source separators (#2106)

     leejet

     authored
     fd43b54View commit details

     Copy full SHA for fd43b54

     Browse repository at this point

   • fix: keep MiniMax-H3 VAE weights resident across temporal chunks (#2103)

     losewayy

     authored
     0963357View commit details

     Copy full SHA for 0963357

     Browse repository at this point

   • feat: enable sdapi latent2RGB previews (#2096)

     Show description for 1da55eb
     

     stduhpf

     and

     leejet

     authored
     1da55ebView commit details

     Copy full SHA for 1da55eb

     Browse repository at this point

   • refactor: unify circular and non-circular tiling (#2105)

     leejet

     authored
     fb8ee71View commit details

     Copy full SHA for fb8ee71

     Browse repository at this point

   • feat: Server: Add support for previews via sdcpp API (#2093)

     Show description for 96f67fe
     

     stduhpf

     and

     leejet

     authored
     96f67feView commit details

     Copy full SHA for 96f67fe

     Browse repository at this point

   • refactor: shorten log levels and move source locations to the end (#2104)

     leejet

     authored
     203a882View commit details

     Copy full SHA for 203a882

     Browse repository at this point

   • fix: correct non-circular tile placement and blending (#2088)

     shawn-mengchen-xu

     authored
     019bcfcView commit details

     Copy full SHA for 019bcfc

     Browse repository at this point

   • fix: correct step count adjustment for custom sigmas list (#2084)

     wbruna

     authored
     c150a6bView commit details

     Copy full SHA for c150a6b

     Browse repository at this point

   • feat: support for merging LoRA weights on model conversion (#2079)

     Show description for f4b5326
     

     wbruna

     and

     leejet

     authored
     f4b5326View commit details

     Copy full SHA for f4b5326

     Browse repository at this point

   • feat: add Z-Image L2P support (#2075)

     aapakhomov

     authored
     bbdcad4View commit details

     Copy full SHA for bbdcad4

     Browse repository at this point

   • fix: avoid f16 overflow in Z-Image quantized matmuls on CUDA (#2074)

     Show description for b6947b8
     

     losewayy

     and

     leejet

     authored
     b6947b8View commit details

     Copy full SHA for b6947b8

     Browse repository at this point

   • perf: avoid redundant permute+cont in non-interleaved RoPE (#2102)

     daniandtheweb

     authored
     9db0db6View commit details

     Copy full SHA for 9db0db6

     Browse repository at this point

   • ci: bundle HIP runtime libraries with Windows ROCm release (#2090)

     harkgill-amd

     authored
     d4bfcd5View commit details

     Copy full SHA for d4bfcd5

     Browse repository at this point
2. Commits on Sep 27, 2026

   • feat: enable hip INT8 tensorwise matmul and convrot (#2071)

     leejet

     authored
     3f8527aView commit details

     Copy full SHA for 3f8527a

     Browse repository at this point

   • fix: keep scaled INT8 convrot matmuls on Vulkan (#2070)

     leejet

     authored
     42ab1c1View commit details

     Copy full SHA for 42ab1c1

     Browse repository at this point

   • fix: update ggml to fix INT8 convrot backend fallback (#2069)

     leejet

     authored
     27f7c43View commit details

     Copy full SHA for 27f7c43

     Browse repository at this point

   • fix: load INT8 convrot LLM embeddings correctly (#2067)

     leejet

     authored
     ede32a6View commit details

     Copy full SHA for ede32a6

     Browse repository at this point

   • sync: update ggml

     leejet

     committed
     47e83d7View commit details

     Copy full SHA for 47e83d7

     Browse repository at this point

   • feat: add Ming-Image Design support (#2063)

     leejet

     authored
     f8890b9View commit details

     Copy full SHA for f8890b9

     Browse repository at this point

   • docs: update hugging face links for MageFlow diffusion and VAE models (#2065)

     akleine

     authored
     61a83b8View commit details

     Copy full SHA for 61a83b8

     Browse repository at this point

   • fix: handle filesystem errors during lora and upscaler cache scan (#2064)

     Yi-111-a

     authored
     9947eebView commit details

     Copy full SHA for 9947eeb

     Browse repository at this point

   • fix: improve GGUF metadata parsing and reader selection (#2062)

     Show description for 168f7b8
     

     losewayy

     and

     leejet

     authored
     168f7b8View commit details

     Copy full SHA for 168f7b8

     Browse repository at this point
3. Commits on Sep 25, 2026

   • refactor: align PixArt weights with upstream layout (#2061)

     leejet

     authored
     2f88688View commit details

     Copy full SHA for 2f88688

     Browse repository at this point

   • refactor: define VAE tile dimensions in image pixels (#2059)

     leejet

     authored
     19bbbcaView commit details

     Copy full SHA for 19bbbca

     Browse repository at this point

   • feat: add PixArt model family support (#2047)

     losewayy

     authored
     39ada08View commit details

     Copy full SHA for 39ada08

     Browse repository at this point

   • fix: load safetensors index shards without recursion (#2058)

     leejet

     authored
     4c3cf75View commit details

     Copy full SHA for 4c3cf75

     Browse repository at this point

   • fix: map Qwen Image 2.1 LoRAs to fused MLP weights (#2057)

     leejet

     authored
     510bccfView commit details

     Copy full SHA for 510bccf

     Browse repository at this point

   • fix: keep the server frontend install out of a parent pnpm workspace (#2056)

     Yi-111-a

     authored
     5e7b291View commit details

     Copy full SHA for 5e7b291

     Browse repository at this point

   • fix: scale Qwen Image 2.1 VAE convolutions (#2054)

     leejet

     authored
     0a9340cView commit details

     Copy full SHA for 0a9340c

     Browse repository at this point
4. Commits on Sep 24, 2026

   • fix: align Qwen Image 2.1 flow schedule with official defaults (#2048)

     leejet

     authored
     b167b94View commit details

     Copy full SHA for b167b94

     Browse repository at this point

   • fix: reserve 128 MiB headroom when selecting monolithic execution (#2046)

     leejet

     authored
     1a2330dView commit details

     Copy full SHA for 1a2330d

     Browse repository at this point

   • feat: add configurable Qwen cache types and early cache scheduling (#2045)

     leejet

     authored
     740c7aeView commit details

     Copy full SHA for 740c7ae

     Browse repository at this point