[tests] implement todos in the modular tests for H3 followup

docs/source/en/api/pipelines/minimax_h3.md

@@ -11,12 +11,6 @@ specific language governing permissions and limitations under the License. -->
 
 # MiniMax-H3
 
-<!-- Remove this note once MiniMax-H3 is part of a diffusers release. -->
-> [!TIP]
-> MiniMax-H3 is not part of a diffusers release yet. Install diffusers from source to use it:
-> `pip install git+https://github.com/huggingface/diffusers.git`
-
-
 MiniMax-H3 generates video and its soundtrack together. A single transformer denoises one packed sequence containing the text conditioning, conditioning media, and target video and audio latents. There is no separate vocoder and no audio post-hoc pass: video and audio come out of the same denoising loop.
 
 You can find the original MiniMax-H3 checkpoints under the [MiniMaxAI](https://huggingface.co/MiniMaxAI) organization.

tests/modular_pipelines/anima/test_modular_pipeline_anima.py

@@ -17,17 +17,13 @@
 
 import numpy as np
 import PIL.Image
-import pytest
 import torch
-from transformers import Qwen2Tokenizer, Qwen3Config, Qwen3Model, T5TokenizerFast
 
 from diffusers import (
     AnimaAutoBlocks,
     AnimaModularPipeline,
     AnimaTextConditioner,
-    AutoencoderKLQwenImage,
     CosmosTransformer3DModel,
-    FlowMatchEulerDiscreteScheduler,
 )
 
 from ...testing_utils import enable_full_determinism, require_peft_backend
@@ -37,12 +33,6 @@
 enable_full_determinism()
 
 
-# TODO: `hf-internal-testing/tiny-anima-modular-pipe` carries no `modular_model_index.json`, so the three tests that
-# build a pipeline from it cannot run. Anima assembles its dummy components in `get_pipeline` instead, which is why
-# nothing noticed until those tests arrived. Publish the tiny repository and drop the overrides below.
-_NO_MODULAR_REPO = "TODO: no tiny Anima modular repository to load from yet."
-
-
 ANIMA_TEXT2IMAGE_WORKFLOWS = {
     "text2image": [
         ("text_encoder", "AnimaTextEncoderStep"),
@@ -72,76 +62,6 @@
 }
 
 
-def get_dummy_components():
-    torch.manual_seed(0)
-    transformer = CosmosTransformer3DModel(
-        in_channels=4,
-        out_channels=4,
-        num_attention_heads=2,
-        attention_head_dim=16,
-        num_layers=2,
-        mlp_ratio=2,
-        text_embed_dim=16,
-        adaln_lora_dim=4,
-        max_size=(4, 32, 32),
-        patch_size=(1, 2, 2),
-        rope_scale=(1.0, 4.0, 4.0),
-        concat_padding_mask=True,
-        extra_pos_embed_type=None,
-    )
-
-    torch.manual_seed(0)
-    vae = AutoencoderKLQwenImage(
-        base_dim=24,
-        z_dim=4,
-        dim_mult=[1, 2, 4],
-        num_res_blocks=1,
-        temperal_downsample=[False, True],
-        latents_mean=[0.0] * 4,
-        latents_std=[1.0] * 4,
-    )
-
-    torch.manual_seed(0)
-    text_conditioner = AnimaTextConditioner(
-        source_dim=16,
-        target_dim=16,
-        model_dim=16,
-        num_layers=2,
-        num_attention_heads=4,
-        target_vocab_size=32128,
-        min_sequence_length=16,
-    )
-
-    torch.manual_seed(0)
-    text_encoder_config = Qwen3Config(
-        vocab_size=152064,
-        hidden_size=16,
-        intermediate_size=32,
-        num_hidden_layers=2,
-        num_attention_heads=4,
-        num_key_value_heads=2,
-        max_position_embeddings=128,
-        rms_norm_eps=1e-6,
-        rope_theta=1000000.0,
-        head_dim=4,
-        attention_bias=False,
-    )
-    text_encoder = Qwen3Model(text_encoder_config).eval()
-    tokenizer = Qwen2Tokenizer.from_pretrained("hf-internal-testing/tiny-random-Qwen2VLForConditionalGeneration")
-    t5_tokenizer = T5TokenizerFast.from_pretrained("hf-internal-testing/tiny-random-t5")
-    scheduler = FlowMatchEulerDiscreteScheduler(shift=3.0)
-
-    return {
-        "transformer": transformer,
-        "vae": vae,
-        "scheduler": scheduler,
-        "text_encoder": text_encoder,
-        "tokenizer": tokenizer,
-        "t5_tokenizer": t5_tokenizer,
-        "text_conditioner": text_conditioner,
-    }
-
-
 def get_dummy_image(height=32, width=32):
     image_array = np.random.randint(0, 256, (height, width, 3), dtype=np.uint8)
     return PIL.Image.fromarray(image_array)
@@ -184,25 +104,6 @@ class TestAnimaModularPipelineFast(ModularPipelineTesterMixin, ModularGuiderTest
     batch_params = frozenset(["prompt", "negative_prompt"])
     expected_workflow_blocks = ANIMA_TEXT2IMAGE_WORKFLOWS
 
-    @pytest.mark.skip(reason=_NO_MODULAR_REPO)
-    def test_from_pretrained_workflow(self):
-        pass
-
-    @pytest.mark.skip(reason=_NO_MODULAR_REPO)
-    def test_load_components_workflow(self):
-        pass
-
-    @pytest.mark.skip(reason=_NO_MODULAR_REPO)
-    def test_unload_components(self):
-        pass
-
-    def get_pipeline(self, components_manager=None, dtype=torch.float32):
-        pipe = self.pipeline_blocks_class().init_pipeline(components_manager=components_manager)
-        pipe.update_components(**get_dummy_components())
-        pipe.to(dtype=dtype)
-        pipe.set_progress_bar_config(disable=None)
-        return pipe
-
     def get_dummy_inputs(self, seed=0):
         generator = torch.Generator(device="cpu").manual_seed(seed)
         return {
@@ -280,25 +181,6 @@ class TestAnimaImg2ImgModularPipelineFast(ModularPipelineTesterMixin):
     batch_params = frozenset(["prompt", "negative_prompt"])
     expected_workflow_blocks = ANIMA_IMG2IMG_WORKFLOWS
 
-    @pytest.mark.skip(reason=_NO_MODULAR_REPO)
-    def test_from_pretrained_workflow(self):
-        pass
-
-    @pytest.mark.skip(reason=_NO_MODULAR_REPO)
-    def test_load_components_workflow(self):
-        pass
-
-    @pytest.mark.skip(reason=_NO_MODULAR_REPO)
-    def test_unload_components(self):
-        pass
-
-    def get_pipeline(self, components_manager=None, torch_dtype=torch.float32):
-        pipe = self.pipeline_blocks_class().init_pipeline(components_manager=components_manager)
-        pipe.update_components(**get_dummy_components())
-        pipe.to(dtype=torch_dtype)
-        pipe.set_progress_bar_config(disable=None)
-        return pipe
-
     def get_dummy_inputs(self, seed=0):
         generator = torch.Generator(device="cpu").manual_seed(seed)
         return {

tests/modular_pipelines/flux2/test_modular_pipeline_flux2_klein.py

@@ -28,14 +28,6 @@
 from ..test_modular_pipelines_common import ModularPipelineTesterMixin
 
 
-# TODO: `hf-internal-testing/tiny-flux2-klein-modular` contradicts itself — `_class_name` is
-# `Flux2KleinModularPipeline` and `is_distilled` is `true`, but `_blocks_class_name` is
-# `Flux2KleinBaseAutoBlocks`. `from_pretrained` honours the latter and builds the base blocks, which declare a
-# `guider` the distilled ones do not, so the pipeline it returns has one component more than these blocks expect.
-# Fix `_blocks_class_name` on the Hub and drop the overrides below.
-_FIXTURE_NAMES_BASE_BLOCKS = "TODO: the tiny repository records `Flux2KleinBaseAutoBlocks` for a distilled checkpoint."
-
-
 FLUX2_KLEIN_WORKFLOWS = {
     "text2image": [
         ("text_encoder", "Flux2KleinTextEncoderStep"),
@@ -60,10 +52,6 @@ class TestFlux2KleinModularPipelineFast(ModularPipelineTesterMixin):
     not_params = frozenset(["negative_prompt"])
     expected_workflow_blocks = FLUX2_KLEIN_WORKFLOWS
 
-    @pytest.mark.skip(reason=_FIXTURE_NAMES_BASE_BLOCKS)
-    def test_from_pretrained_workflow(self):
-        pass
-
     def get_dummy_inputs(self, seed=0):
         generator = self.get_generator(seed)
         inputs = {
@@ -110,10 +98,6 @@ class TestFlux2KleinImageConditionedModularPipelineFast(ModularPipelineTesterMix
     not_params = frozenset(["negative_prompt"])
     expected_workflow_blocks = FLUX2_KLEIN_IMAGE_CONDITIONED_WORKFLOWS
 
-    @pytest.mark.skip(reason=_FIXTURE_NAMES_BASE_BLOCKS)
-    def test_from_pretrained_workflow(self):
-        pass
-
     def get_dummy_inputs(self, seed=0):
         generator = self.get_generator(seed)
         inputs = {

tests/modular_pipelines/qwen/test_modular_pipeline_qwenimage.py

@@ -154,12 +154,6 @@ def test_inference_batch_single_identical(self):
         super().test_inference_batch_single_identical(expected_max_diff=5e-4)
 
 
-# TODO: `hf-internal-testing/tiny-qwenimage-edit-modular` records `_class_name: QwenImageModularPipeline` and
-# `_blocks_class_name: SequentialPipelineBlocks`, so `from_pretrained` falls back to `QwenImageAutoBlocks`, which
-# declares no `image_conditioned` workflow. Point the repository at the Edit classes and drop the overrides below.
-_FIXTURE_NAMES_BASE_PIPELINE = "TODO: the tiny repository records `QwenImageModularPipeline` for an Edit checkpoint."
-
-
 QWEN_IMAGE_EDIT_WORKFLOWS = {
     "image_conditioned": [
         ("text_encoder.resize", "QwenImageEditResizeStep"),
@@ -207,14 +201,6 @@ class TestQwenImageEditModularPipelineFast(ModularPipelineTesterMixin, ModularGu
     batch_params = frozenset(["prompt", "negative_prompt", "image", "mask_image"])
     expected_workflow_blocks = QWEN_IMAGE_EDIT_WORKFLOWS
 
-    @pytest.mark.skip(reason=_FIXTURE_NAMES_BASE_PIPELINE)
-    def test_from_pretrained_workflow(self):
-        pass
-
-    @pytest.mark.skip(reason=_FIXTURE_NAMES_BASE_PIPELINE)
-    def test_load_components_workflow(self):
-        pass
-
     def get_dummy_inputs(self):
         generator = self.get_generator()
         inputs = {