All pull requests

Pull requests

• Add GGML MoEExperts (Ming-Image) + fix F32 tokenizer_json tensors

  #486·bbirdkt opened Sep 25, 2026·

  ·

• Load mmproj vision tower for qwen3vl text encoders (fixes Qwen-Image-2.1 12288 error)

  #485·pottokao-dotcom opened Sep 23, 2026·

  ·

• Fix Ming-Image GGUF packed QKV loading

  #484·RealRebelAI opened Sep 23, 2026·

  ·
  2

• Add Qwen-Image 2.1 architecture detection (sd.cpp GGUF exports fail with 'Unknown model architecture!')

  #483·NikolayGusev-astra opened Sep 21, 2026·

  ·

• Fix MiniMax H3 GGUF model and encoder loading

  #481·Gingertrout opened Sep 18, 2026·

  ·

• Add MiniMax H3 architecture detection for metadata-less GGUF files

  #480·uppet opened Sep 8, 2026·

  ·

• Fix KeyError: 'default_num_special_tokens' when recreating Tekken tok…

  #479·bigdatamark opened Sep 5, 2026·

  ·

• Fix Qwen3-VL (qwen3vl) mmproj not loaded: Krea2 / Qwen3-VL CLIP falls back to QWEN3_4B

  #478·allanmeng opened Sep 2, 2026·

  ·

• Add minimax_h3 to IMG_ARCH_LIST

  #476·GabrielSalerno03 opened Aug 20, 2026·

  ·

• MiniMax H3 mmproj support for Qwen3 VL 32B GGUFs

  #473·Nif00 opened Aug 5, 2026·

  ·
  1

• Cache device-side constants in dequant kernels

  #472·lanslote opened Aug 3, 2026·

  ·
  1

• Fix QK-norm .scale/.weight key mismatch on Flux-compat GGUF loads (silent NaN/black image)

  #470·ayanoby opened Aug 2, 2026·

  ·

• Ensure force_patch_weights is set when partially loading/unloading models

  #469·blepping opened Jul 26, 2026Contributor·

  ·

• feat(ops): add GGMLTensor.dequantize() for generic core cast path

  #468·Log-Dog012 opened Jul 23, 2026·

  ·
  3

• Fix loading of BF16 bare nn.Parameters (LTX-2 learnable_registers)

  #467·hc20k opened Jul 16, 2026·

  ·

• Handle WeightAdapters in ops.move_patch_to_device

  #461·blepping opened Jun 16, 2026Contributor·

  ·

• Add Ideogram 4 architecture support

  #460·DanrisiUA opened Jun 12, 2026·

  ·
  5

• Add Ideogram support and improve BF16 dequantization handling

  #459·molbal opened Jun 9, 2026·

  ·
  8

• Report a float compute dtype for quantized GGMLTensor

  #456·fxd0h opened Jun 7, 2026·

  ·

• Add Ideogram-4 diffusion model architecture support

  #455·fxd0h opened Jun 7, 2026·

  ·

• nodes: skip self.model.to(device_to) walk in unpatch_model on mmap'd weights

  #445·Booyaka101 opened May 10, 2026·

  ·
  1

  6

• Add mistral3 architecture support for Ministral 3B

  #440·insecure-erasure opened Apr 18, 2026·

  ·
  1

  2

• added qwen3.5 support

  #438·TheJoshCode opened Apr 15, 2026·

  ·

• Add basic Ministral 3 3B support (Ernie Image)

  #436·deinferno opened Apr 14, 2026·

  ·
  7

• Update loader.py, fix typo

  #434·RubenGarcia opened Apr 6, 2026·

  ·