Repository navigation

Add Boogu-Image generation, editing, and turbo pipelines - #14040

Open

Boogu-Team wants to merge 28 commits into

huggingface:mainfrom

Boogu-Team:feat/integrate-boogu

Open

Add Boogu-Image generation, editing, and turbo pipelines#14040
Boogu-Team wants to merge 28 commits into
huggingface:mainfrom
Boogu-Team:feat/integrate-boogu

Commits on Jun 18, 2026
- Add Boogu-Image generation and editing pipeline
  
  Boogu-Team
  and
  claude
  committed
  c09bbcb
Commits on Jun 22, 2026
- Boogu: remove TaylorSeer cache, keep official scheduler + TeaCache
  
  Boogu-Team
  and
  claude
  committed
  4da1be8
- Use official FlowMatchEulerDiscreteScheduler for Boogu via time-shift adapter
  
  Boogu-Team
  and
  claude
  committed
  f765d13
- Slim Boogu pipeline: inline scheduler adapter, drop LoRA mixin and rewriter
  
  Boogu-Team
  and
  claude
  committed
  614f3a0
- examples/boogu: add negative_instruction to edit examples
  
  Boogu-Team
  and
  claude
  committed
  74a0fad
- Boogu: mechanical cleanup for upstream PR (dead code, logging, conventions)
  
  Boogu-Team
  and
  claude
  committed
  1741e80
- Boogu: remove prompt-tuning subsystem
  
  Boogu-Team
  and
  claude
  committed
  059451f
- Boogu: drop except-Exception fallback in instruction encoding
  
  Boogu-Team
  and
  claude
  committed
  21cf345
- Boogu: route attention through dispatch_attention_fn
  
  Boogu-Team
  and
  claude
  committed
  ecc389b
- Boogu: make BooguImageTurboPipeline a standalone pipeline
  
  Boogu-Team
  and
  claude
  committed
  ee3738f
- Boogu: second-pass cleanup for upstream PR
  
  Boogu-Team
  and
  claude
  committed
  aa1252f
- Boogu: apply ruff format
  
  Boogu-Team
  and
  claude
  committed
  8952e6d
- Boogu: drop stale ruff ignore for removed static-skills file
  
  Boogu-Team
  and
  claude
  committed
  150ad3b
Commits on Jun 23, 2026
- Boogu: drop triton fused RMSNorm, use torch.nn.RMSNorm
  
  Boogu-Team
  and
  claude
  committed
  9e672c2
- Boogu: consolidate model into a single file (single-file convention)
  
  Boogu-Team
  and
  claude
  committed
  d202a23
- Boogu: use base-class device + offload management
  
  Boogu-Team
  and
  claude
  committed
  5cef903
Commits on Jun 25, 2026
- Boogu: drop weight-init, follow lazy __init__ convention
  
  Boogu-Team
  and
  claude
  committed
  05cc899
- Boogu: rename Lumina-derived layers to BooguImage*
  
  Boogu-Team
  and
  claude
  committed
  a597eed
- Boogu: collapse RoPE namespace class into a get_freqs_cis function
  
  Boogu-Team
  and
  claude
  committed
  36298d3
- Boogu: remove TeaCache, defer to CacheMixin
  
  Boogu-Team
  and
  claude
  committed
  c9a6b5d
- Boogu: compute time-shift mu via calculate_shift
  
  Boogu-Team
  and
  claude
  committed
  662e6ca
- Boogu: introduce BooguImageAttention module, dispatch via processors
  
  Boogu-Team
  and
  claude
  committed
  2977b2d
- Boogu: remove boosted orthogonal guidance, keep standard CFG
  
  Boogu-Team
  and
  claude
  committed
  3b56e27
Commits on Jul 9, 2026
- Boogu: swiglu cleanup, drop redundant mask var, move output class
  
  Boogu-Team
  and
  claude
  committed
  9918a63
- Boogu: inline patch-embed + refiner layer calls into forward
  
  Boogu-Team
  and
  claude
  committed
  96fd58b
- Boogu: accept only None / list / list-of-list for input_images, else raise
  
  Boogu-Team
  and
  claude
  committed
  b5fe1de
- Boogu: install timesteps via set_timesteps(sigmas=), feed t/num_train_timesteps
  
  Boogu-Team
  and
  claude
  committed
  61bb010
- Boogu: drop redundant nested torch.no_grad() in _get_instruction_feature_embeds
  
  Boogu-Team
  and
  claude
  committed
  b8be806

Navigation Menu

Repository navigation

Add Boogu-Image generation, editing, and turbo pipelines - #14040

Add Boogu-Image generation, editing, and turbo pipelines#14040
Boogu-Team wants to merge 28 commits into
huggingface:mainfrom
Boogu-Team:feat/integrate-boogu

Commits on Jun 18, 2026

Add Boogu-Image generation and editing pipeline

Commits on Jun 22, 2026

Boogu: remove TaylorSeer cache, keep official scheduler + TeaCache

Use official FlowMatchEulerDiscreteScheduler for Boogu via time-shift adapter

Slim Boogu pipeline: inline scheduler adapter, drop LoRA mixin and rewriter

examples/boogu: add negative_instruction to edit examples

Boogu: mechanical cleanup for upstream PR (dead code, logging, conventions)

Boogu: remove prompt-tuning subsystem

Boogu: drop except-Exception fallback in instruction encoding

Boogu: route attention through dispatch_attention_fn

Boogu: make BooguImageTurboPipeline a standalone pipeline

Boogu: second-pass cleanup for upstream PR

Boogu: apply ruff format

Boogu: drop stale ruff ignore for removed static-skills file

Commits on Jun 23, 2026

Boogu: drop triton fused RMSNorm, use torch.nn.RMSNorm

Boogu: consolidate model into a single file (single-file convention)

Boogu: use base-class device + offload management

Commits on Jun 25, 2026

Boogu: drop weight-init, follow lazy init convention

Boogu: rename Lumina-derived layers to BooguImage*

Boogu: collapse RoPE namespace class into a get_freqs_cis function

Boogu: remove TeaCache, defer to CacheMixin

Boogu: compute time-shift mu via calculate_shift

Boogu: introduce BooguImageAttention module, dispatch via processors

Boogu: remove boosted orthogonal guidance, keep standard CFG

Commits on Jul 9, 2026

Boogu: swiglu cleanup, drop redundant mask var, move output class

Boogu: inline patch-embed + refiner layer calls into forward

Boogu: accept only None / list / list-of-list for input_images, else raise

Boogu: install timesteps via set_timesteps(sigmas=), feed t/num_train_timesteps

Boogu: drop redundant nested torch.no_grad() in _get_instruction_feature_embeds