* Community Pipeline: Add z-image differential img2img
* add pipeline for z-image differential img2img diffusion examples : run make style , make quality, and fix white spaces in example doc string.
---------
Co-authored-by: r4inm4ker <jefri.yeh@gmail.com>
* fix torchao quantizer for new torchao versions
Summary:
`torchao==0.16.0` (not yet released) has some bc-breaking changes, this
PR fixes the diffusers repo with those changes. Specifics on the
changes:
1. `UInt4Tensor` is removed: https://github.com/pytorch/ao/pull/3536
2. old float8 tensors v1 are removed: https://github.com/pytorch/ao/pull/3510
In this PR:
1. move the logger variable up (not sure why it was in the middle of the
file before) to get better error messages
2. gate the old torchao objects by torchao version
Test Plan:
import diffusers objects with new versions of torchao works:
```bash
> python -c "import torchao; print(torchao.__version__); from diffusers import StableDiffusionPipeline"
0.16.0.dev20251229+cu129
```
Reviewers:
Subscribers:
Tasks:
Tags:
* Apply style fixes
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
* Add z-image-omni-base implementation
* Merged into one transformer for Z-Image.
* Fix bugs for controlnet after merging the main branch new feature.
* Fix for auto_pipeline, Add Styling.
* Refactor noise handling and modulation
- Add select_per_token function for per-token value selection
- Separate adaptive modulation logic
- Cleanify t_noisy/clean variable naming
- Move image_noise_mask handler from forward to pipeline
* Styling & Formatting.
* Rewrite code with more non-forward func & clean forward.
1.Change to one forward with shorter code with omni code (None).
2.Split out non-forward funcs: _build_unified_sequence, _prepare_sequence, patchify, pad.
* Styling & Formatting.
* Manual check fix-copies in controlnet, Add select_per_token, _patchify_image, _pad_with_ids; Styling.
* Add Import in pipeline __init__.py.
---------
Co-authored-by: Jerry Qilong Wu <xinglong.wql@alibaba-inc.com>
Co-authored-by: YiYi Xu <yixu310@gmail.com>
Use `T5Tokenizer` instead of `MT5Tokenizer`
Given that the `MT5Tokenizer` in `transformers` is just a "re-export" of
`T5Tokenizer` as per
https://github.com/huggingface/transformers/blob/v4.57.3/src/transformers/models/mt5/tokenization_mt5.py
)on latest available stable Transformers i.e., v4.57.3), this commit
updates the imports to point to `T5Tokenizer` instead, so that those
still work with Transformers v5.0.0rc0 onwards.
* move node registry to mellon
* up
* fix
* modula rpipeline update: filter out none for input_names, fix default blocks for pipe.init() and allow user pass additional kwargs_type in a dict
* qwen modular refactor, unpack before decode
* update mellon node config, adding* to required_inputs and required_model_inputs
* modularpipeline.from_pretrained: error out if no config found
* add a component_names property to modular blocks to be consistent!
* flux image_encoder -> vae_encoder
* controlnet_bundle
* refator MellonNodeConfig MellonPipelineConfig
* refactor & simplify mellon utils
* vae_image_encoder -> vae_encoder
* mellon config save keep key order
* style + copies
* add kwargs input for zimage
* cosmos predict2.5 base: convert chkpt & pipeline
- New scheduler: scheduling_flow_unipc_multistep.py
- Changes to TransformerCosmos for text embeddings via crossattn_proj
* scheduler cleanup
* simplify inference pipeline
* cleanup scheduler + tests
* Basic tests for flow unipc
* working b2b inference
* Rename everything
* Tests for pipeline present, but not working (predict2 also not working)
* docstring update
* wrapper pipelines + make style
* remove unnecessary files
* UniPCMultistep: support use_karras_sigmas=True and use_flow_sigmas=True
* use UniPCMultistepScheduler + fix tests for pipeline
* Remove FlowUniPCMultistepScheduler
* UniPCMultistepScheduler for use_flow_sigmas=True & use_karras_sigmas=True
* num_inference_steps=36 due to bug in scheduler used by predict2.5
* Address comments
* make style + make fix-copies
* fix tests + remove references to old pipelines
* address comments
* add revision in from_pretrained call
* fix tests
* extend TorchAoTest::test_model_memory_usage to other platform
Signe-off-by: Wang, Yi <yi.a.wang@inel.com>
* add some comments
Signed-off-by: Wang, Yi <yi.a.wang@intel.com>
---------
Signed-off-by: Wang, Yi <yi.a.wang@intel.com>
fix pytest tests/pipelines/pixart_sigma/test_pixart.py::PixArtSigmaPipelineIntegrationTests::test_pixart_512 in xpu
Signed-off-by: Wang, Yi <yi.a.wang@intel.com>
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* Add ZImageImg2ImgPipeline
Updated the pipeline structure to include ZImageImg2ImgPipeline
alongside ZImagePipeline.
Implemented the ZImageImg2ImgPipeline class for image-to-image
transformations, including necessary methods for
encoding prompts, preparing latents, and denoising.
Enhanced the auto_pipeline to map the new ZImageImg2ImgPipeline
for image generation tasks.
Added unit tests for ZImageImg2ImgPipeline to ensure
functionality and performance.
Updated dummy objects to include ZImageImg2ImgPipeline for
testing purposes.
* Address review comments for ZImageImg2ImgPipeline
- Add `# Copied from` annotations to encode_prompt and _encode_prompt
- Add ZImagePipeline to auto_pipeline.py for AutoPipeline support
* Add ZImage pipeline documentation
---------
Co-authored-by: YiYi Xu <yixu310@gmail.com>
Co-authored-by: Álvaro Somoza <asomoza@users.noreply.github.com>
* feat: Add `flow_prediction` to `prediction_type`, introduce `use_flow_sigmas`, `flow_shift`, `use_dynamic_shifting`, and `time_shift_type` parameters, and refine type hints for various arguments.
* style: reformat argument wrapping in `_convert_to_beta` and `index_for_timestep` method signatures.
* fix: group offloading to support standalone computational layers in block-level offloading
* test: for models with standalone and deeply nested layers in block-level offloading
* feat: support for block-level offloading in group offloading config
* fix: group offload block modules to AutoencoderKL and AutoencoderKLWan
* fix: update group offloading tests to use AutoencoderKL and adjust input dimensions
* refactor: streamline block offloading logic
* Apply style fixes
* update tests
* update
* fix for failing tests
* clean up
* revert to use skip_keys
* clean up
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>
