Fanli Lin

make `pipelines` tests device-agnostic (part2)

1

## What does this PR do? Followed by [this ](https://github.com/huggingface/diffusers/pull/9399) PR

make `pipelines` tests device-agnostic (part1)

16

## What does this PR do? Below are some evidences: ```bash PASSED tests/pipelines/animatediff/test_animatediff.py::AnimateDiffPipelineFastTests::test_to_device PASSED tests/pipelines/amused/test_amused.py::AmusedPipelineFastTests::test_model_cpu_offload_forward_pass PASSED tests/pipelines/amused/test_amused.py::AmusedPipelineFastTests::test_cpu_offload_ forward_pass_twice PASSED tests/pipelines/amused/test_amused_inpaint.py::AmusedInpaintPipelineFastTests::te st_to_device ``` @yiyixuxu

make `single_file` tests device-agnostic

1

# What does this PR do? Fixes # (issue) ## Before submitting - [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks...

add the missing xpu for local sgd

2

## What does this PR do? As the tile suggests, local sgd also works on XPU.

take `torch.nn.Module` model into account when moving to device

1

## What does this PR do? Last time in #3133, I introduced the check to only move the model when the model is on "cpu". But this doesn't take the...

[docs] add xpu part and fix bug in `torchrun`

2

## What does this PR do? This PR adds xpu and use `nnodes` instead of `num_machines` for torchrun. @muellerzr

can "tests" be used instead of "test" for test folder?

3

Hi, all other huggingface libraries use "tests" for the test folder instead of "test", e.g. https://github.com/huggingface/transformers, https://github.com/huggingface/accelerate. If it is okay for you, I can submit a PR for the...

`datasets.map(..., num_proc=4)` multi-processing fails

1

The following code fails in python 3.11+ ```python tokenized_datasets = datasets.map(tokenize_function, batched=True, num_proc=4, remove_columns=["text"]) ``` Error log: ```bash Traceback (most recent call last): File "/usr/local/lib/python3.12/dist-packages/multiprocess/process.py", line 315, in _bootstrap self.run()...

[Feat] Support XPU Backend in vLLM-Omni

6

## Purpose This PR introduces foundational support for the XPU backend in vLLM-Omni, enabling inference and serving of omni-modality models on Intel XPU devices. Changes include: - XPU worker&model runner...

## Purpose Added documentation for the image-to-video offline inference example, covering both Wan2.2-I2V-A14B-Diffusers (MoE) and Wan2.2-TI2V-5B-Diffusers (unified) models. But during testing, I found the TI2V model is broken. Issue created:...