[CI] Refactor Wan Model Tests #13082

DN6 · 2026-02-04T12:58:35Z

What does this PR do?

Update Wan tests with new format

Fixes # (issue)

Before submitting

This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
Did you read the contributor guideline?
Did you read our philosophy doc (important for complex PRs)?
Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
Did you make sure to update the documentation with your changes? Here are the
documentation guidelines, and
here are tips on formatting docstrings.
Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

yiyixuxu

really nice thanks!
should we start to add a guide for contributor some where, maybe https://huggingface.co/docs/diffusers/main/en/conceptual/contribution

dg845

Thanks! I see there are two Wan model related failures from the CI:

tests/models/transformers/test_models_transformer_wan_animate.py::TestWanAnimateTransformer3DAttention::test_fuse_unfuse_qkv_projections
tests/models/transformers/test_models_transformer_wan_vace.py::TestWanVACETransformer3DAttention::test_fuse_unfuse_qkv_projections

If I try to run the new Wan tests locally, for example with

pytest tests/models/transformers/test_models_transformer_wan.py

I get some more test failures:

tests/models/transformers/test_models_transformer_wan.py::TestWanTransformer3D
- test_keep_in_fp32_modules
- test_from_save_pretrained_dtype_inference[fp16,bf16]
tests/models/transformers/test_models_transformer_wan.py::TestWanTransformer3DGGUF
- test_gguf_quantization_inference
- test_gguf_keep_modules_in_fp32
- test_gguf_quantization_dtype_assignment
- test_gguf_quantization_lora_inference
- test_gguf_dequantize
- test_gguf_quantized_layers
tests/models/transformers/test_models_transformer_wan.py::TestWanTransformer3DGGUFCompile
- test_gguf_torch_compile
- test_gguf_torch_compile_with_group_offload

Are these test failures expected?

HuggingFaceDocBuilderDev · 2026-02-06T09:39:32Z

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

DN6 · 2026-02-06T11:07:49Z

Thanks for flagging @dg845. I've fixed the test issues. There are some GGUF related fixes that should probably go in a different PR (will handle that later)

dg845

Thanks!

src/diffusers/models/transformers/transformer_wan.py

sayakpaul

Thanks, I left some comments!

sayakpaul · 2026-02-08T06:07:44Z

src/diffusers/models/transformers/transformer_wan_animate.py

        self.inner_dim = dim_head * heads
        self.heads = heads
-        self.cross_attention_head_dim = cross_attention_dim_head
+        self.cross_attention_dim_head = cross_attention_dim_head


Same as above. Would be nice if you could explain these changes? Were these flagged by the newly written test suite?

This is just to keep the naming convention consistent

tests/models/testing_utils/common.py

sayakpaul · 2026-02-08T06:09:44Z

tests/models/testing_utils/quantization.py

-        # Get model dtype from first parameter
-        model_dtype = next(model_quantized.parameters()).dtype
-
        inputs = self.get_dummy_inputs()
-        # Cast inputs to model dtype
-        inputs = {
-            k: v.to(model_dtype) if isinstance(v, torch.Tensor) and v.is_floating_point() else v
-            for k, v in inputs.items()
-        }


Why remove them?

Casting here is brittle because it's based on model_dtype which we get from model_dtype = next(model_quantized.parameters()).dtype. This can lead to different dtypes across different models and different quantization schemes. e.g With Flux + GGUF the test passes because the parameter dtype is the same the input dtype (bfloat16). However with Wan it fails because the parameter dtype is int8.

Makes sense. But does it affect the existing Flux tests?

Casting here is brittle because it's based on model_dtype which we get from model_dtype = next(model_quantized.parameters()).dtype

I wonder if using .dtype on a model subclassed from ModelMixin would alleviate this problem because dtype implementation is quite elaborate:

diffusers/src/diffusers/models/modeling_utils.py

Line 155 in 5bf248d

def get_parameter_dtype(parameter: torch.nn.Module) -> torch.dtype:

I've added a torch_dtype property to the quantization tests and we cast the inputs directly in get_dummy_inputs. Think it's more clear this way

Flux TorchAO and BnB tests will fail with this change, but I'll update the Flux2 PR to include fixes to address the change in this test.

Sounds good. Thanks!

tests/models/testing_utils/quantization.py

sayakpaul · 2026-02-08T06:11:31Z

tests/models/transformers/test_models_transformer_wan.py

 # See the License for the specific language governing permissions and
 # limitations under the License.

-import unittest


I am guessing the changes under tests/models/transformers/ were all auto-generated?

sayakpaul · 2026-02-08T06:12:59Z

tests/models/transformers/test_models_transformer_wan_vace.py

+class TestWanVACETransformer3DCompile(WanVACETransformer3DTesterConfig, TorchCompileTesterMixin):
+    """Torch compile tests for Wan VACE Transformer 3D."""
+
+    def test_torch_compile_repeated_blocks(self):


I think we can further simplify this test by letting users pass a recompile_limit. I will open a PR.

update

e8a3ef8

DN6 requested review from dg845, sayakpaul and yiyixuxu February 4, 2026 12:58

yiyixuxu approved these changes Feb 4, 2026

View reviewed changes

dg845 reviewed Feb 5, 2026

View reviewed changes

update

42cd24c

DN6 added 3 commits February 6, 2026 11:38

update

f12a9fd

update

1886198

update

5776aed

update

41a26b7

dg845 approved these changes Feb 7, 2026

View reviewed changes

sayakpaul reviewed Feb 8, 2026

View reviewed changes

src/diffusers/models/transformers/transformer_wan.py Show resolved Hide resolved

sayakpaul reviewed Feb 8, 2026

View reviewed changes

DN6 added 2 commits February 10, 2026 09:45

update

9337364

update

e99d30d

DN6 merged commit c3a4cd1 into main Feb 11, 2026
11 of 12 checks passed

[CI] Refactor Wan Model Tests #13082

[CI] Refactor Wan Model Tests #13082

Conversation

DN6 commented Feb 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Before submitting

Who can review?

Uh oh!

yiyixuxu left a comment

Choose a reason for hiding this comment

Uh oh!

dg845 left a comment

Choose a reason for hiding this comment

Uh oh!

HuggingFaceDocBuilderDev commented Feb 6, 2026

Uh oh!

DN6 commented Feb 6, 2026

Uh oh!

dg845 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

sayakpaul left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

DN6 commented Feb 4, 2026 •

edited

Loading