All pull requests

Pull requests

• fix(offload): move blocks to CPU before deepcopy in CPUOffloadManager to avoid deadlock

  #951·LacklusterOpsec opened Aug 26, 2026·

  ·

• fix: from_linear dereferences linear.weight even when torch_dtype is passed

  #949·JoaoZaokk opened Aug 18, 2026·

  ·

• feat: add Krea 2 W4A4 support

  #947·sztlink opened Jul 20, 2026·

  ·

• Fix int32 overflows in W4A4 kernels at high resolution

  #944·danielwoz opened Jul 10, 2026·

  ·

• Support lazy-initialized Linear modules in SVDQW4A4Linear.from_linear

  #929·kj863257 opened Apr 7, 2026·

  ·

• feat:support new model:chroma

  #928·tonera opened Apr 3, 2026·

  ·
  1

• feat: add Qwen encoder runtimes (Qwen2.5-VL + Qwen3) for Nunchaku

  #927·tonera opened Mar 31, 2026·

  ·

• feat: add FLUX.2 transformer runtime support (nunchaku)

  #926·tonera opened Mar 31, 2026·

  ·
  6

• fix: use tuple for multidimensional indexing in lora/flux/utils.py

  #921·j2gg0s opened Mar 19, 2026·

  ·

• feat: add FBCache support for HunyuanVideo

  #918·Bluear7878 opened Mar 9, 2026Contributor·

  ·

• Fix: UserWarning: Using a non-tuple sequence for multidimensional indexing is deprecated and will be changed in pytorch 2.9

  #914·angerrp opened Mar 6, 2026·

  ·

• Remove redundant IP-adapter computation & vectorize add_kernel

  #912·Bluear7878 opened Mar 4, 2026Contributor·

  ·

• feat(lora): implement fast LoRA switching for instant variant switching

  #897·sqhuang opened Jan 29, 2026·

  ·
  1

• [Qwen] Validate encoder_hidden_states for text sequence length

  #887·kashif opened Jan 19, 2026·

  ·
  1

• feat(zimage): add LoRA support for NunchakuZImageTransformer2DModel

  #866·sqhuang opened Jan 5, 2026·

  ·

• feat: Enable deterministic LoRA activation reduction using fixed-point arithmetic

  #820·lantudou opened Dec 17, 2025·

  ·
  1

• Add diverse examples for Cache-dit in Flux.1 and Qwen models regarding #763

  #810·Bluear7878 opened Nov 26, 2025Contributor·

  ·

• feat: NunchakuFluxTransformer2DModelV2 supports lora

  #792·devgdovg opened Oct 31, 2025Contributor·

  ·
  1

• Add SDXL FBCache Support

  #787·Bluear7878 opened Oct 28, 2025Contributor·

  ·
  1

• Remove unexpected code in compose_lora to resolve bugs

  #566·iceload123 opened Jul 25, 2025·

  ·
  1

• add filter_prefix and del_filter_prefix

  #472·SongDI911 opened Jun 16, 2025·

  ·
  6

• fix: fix the LoRA loading issue in #265

  #384·StarMingGao opened May 21, 2025·

  ·

• Add cog.yaml and predict.py for usage with Replicate and Docker

  #39·voodoohop opened Dec 1, 2024·

  ·
  3