Merge branch 'main' into fix/z-image-redundant-timestep

Author: rootonchair

.github/workflows/build_documentation.yml

@@ -25,6 +25,7 @@ jobs:
       notebook_folder: diffusers_doc
       languages: en ko zh ja pt
       custom_container: diffusers/diffusers-doc-builder
+      pre_command: uv pip uninstall transformers huggingface_hub && UV_PRERELEASE=allow uv pip install -U transformers@git+https://github.com/huggingface/transformers.git
     secrets:
       token: ${{ secrets.HUGGINGFACE_PUSH }}
       hf_token: ${{ secrets.HF_DOC_BUILD_PUSH }}

.github/workflows/build_pr_documentation.yml

@@ -50,3 +50,4 @@ jobs:
       package: diffusers
       languages: en ko zh ja pt
       custom_container: diffusers/diffusers-doc-builder
+      pre_command: uv pip uninstall transformers huggingface_hub && UV_PRERELEASE=allow uv pip install -U transformers@git+https://github.com/huggingface/transformers.git

.github/workflows/pr_tests.yml

@@ -194,6 +194,8 @@ jobs:
     - name: Install dependencies
       run: |
         uv pip install -e ".[quality]"
+        uv pip uninstall transformers huggingface_hub && UV_PRERELEASE=allow uv pip install -U transformers@git+https://github.com/huggingface/transformers.git
+        uv pip uninstall tokenizers && uv pip install "tokenizers<=0.23.0"
 
     - name: Environment
       run: |

src/diffusers/__init__.py

@@ -22,6 +22,7 @@
     is_torchao_available,
     is_torchsde_available,
     is_transformers_available,
+    is_transformers_flax_compatible,
     is_transformers_version,
 )
 
@@ -861,7 +862,6 @@
     _import_structure["models.modeling_flax_utils"] = ["FlaxModelMixin"]
     _import_structure["models.unets.unet_2d_condition_flax"] = ["FlaxUNet2DConditionModel"]
     _import_structure["models.vae_flax"] = ["FlaxAutoencoderKL"]
-    _import_structure["pipelines"].extend(["FlaxDiffusionPipeline"])
     _import_structure["schedulers"].extend(
         [
             "FlaxDDIMScheduler",
@@ -878,7 +878,7 @@
 
 
 try:
-    if not (is_flax_available() and is_transformers_available()):
+    if not (is_flax_available() and is_transformers_available() and is_transformers_flax_compatible()):
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
     from .utils import dummy_flax_and_transformers_objects  # noqa F403
@@ -891,6 +891,7 @@
 else:
     _import_structure["pipelines"].extend(
         [
+            "FlaxDiffusionPipeline",
             "FlaxStableDiffusionControlNetPipeline",
             "FlaxStableDiffusionImg2ImgPipeline",
             "FlaxStableDiffusionInpaintPipeline",
@@ -1620,7 +1621,6 @@
         from .models.modeling_flax_utils import FlaxModelMixin
         from .models.unets.unet_2d_condition_flax import FlaxUNet2DConditionModel
         from .models.vae_flax import FlaxAutoencoderKL
-        from .pipelines import FlaxDiffusionPipeline
         from .schedulers import (
             FlaxDDIMScheduler,
             FlaxDDPMScheduler,
@@ -1634,12 +1634,13 @@
         )
 
     try:
-        if not (is_flax_available() and is_transformers_available()):
+        if not (is_flax_available() and is_transformers_available() and is_transformers_flax_compatible()):
             raise OptionalDependencyNotAvailable()
     except OptionalDependencyNotAvailable:
         from .utils.dummy_flax_and_transformers_objects import *  # noqa F403
     else:
         from .pipelines import (
+            FlaxDiffusionPipeline,
             FlaxStableDiffusionControlNetPipeline,
             FlaxStableDiffusionImg2ImgPipeline,
             FlaxStableDiffusionInpaintPipeline,

src/diffusers/modular_pipelines/ernie_image/encoders.py

@@ -15,16 +15,23 @@
 import json
 
 import torch
-from transformers import AutoModel, AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoTokenizer, Mistral3Model
 
 from ...configuration_utils import FrozenDict
 from ...guiders import ClassifierFreeGuidance
 from ...utils import logging
+from ...utils.import_utils import is_transformers_version
 from ..modular_pipeline import ModularPipelineBlocks, PipelineState
 from ..modular_pipeline_utils import ComponentSpec, InputParam, OutputParam
 from .modular_pipeline import ErnieImageModularPipeline
 
 
+if is_transformers_version("<", "5.0.0"):
+    raise ImportError("`ErnieImageModularPipeline` requires `transformers>=5.0.0` for `Ministral3ForCausalLM`.")
+
+from transformers import Ministral3ForCausalLM  # noqa: E402
+
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
 
@@ -38,7 +45,7 @@ def description(self) -> str:
     @property
     def expected_components(self) -> list[ComponentSpec]:
         return [
-            ComponentSpec("pe", AutoModelForCausalLM),
+            ComponentSpec("pe", Ministral3ForCausalLM),
             ComponentSpec("pe_tokenizer", AutoTokenizer),
         ]
 
@@ -83,7 +90,7 @@ def intermediate_outputs(self) -> list[OutputParam]:
 
     @staticmethod
     def _enhance_prompt(
-        pe: AutoModelForCausalLM,
+        pe: Ministral3ForCausalLM,
         pe_tokenizer: AutoTokenizer,
         prompt: str,
         device: torch.device,
@@ -160,7 +167,7 @@ def description(self) -> str:
     @property
     def expected_components(self) -> list[ComponentSpec]:
         return [
-            ComponentSpec("text_encoder", AutoModel),
+            ComponentSpec("text_encoder", Mistral3Model),
             ComponentSpec("tokenizer", AutoTokenizer),
             ComponentSpec(
                 "guider",
@@ -200,7 +207,7 @@ def intermediate_outputs(self) -> list[OutputParam]:
 
     @staticmethod
     def _encode(
-        text_encoder: AutoModel,
+        text_encoder: Mistral3Model,
         tokenizer: AutoTokenizer,
         prompt: list[str],
         device: torch.device,

src/diffusers/pipelines/__init__.py

@@ -5,7 +5,6 @@
     OptionalDependencyNotAvailable,
     _LazyModule,
     get_objects_from_module,
-    is_flax_available,
     is_librosa_available,
     is_note_seq_available,
     is_onnx_available,
@@ -14,6 +13,7 @@
     is_torch_available,
     is_torch_npu_available,
     is_transformers_available,
+    is_transformers_flax_compatible,
     is_transformers_version,
 )
 
@@ -504,7 +504,7 @@
     _import_structure["consisid"] = ["ConsisIDPipeline"]
 
 try:
-    if not is_flax_available():
+    if not is_transformers_flax_compatible():
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
     from ..utils import dummy_flax_objects  # noqa F403
@@ -513,7 +513,7 @@
 else:
     _import_structure["pipeline_flax_utils"] = ["FlaxDiffusionPipeline"]
 try:
-    if not (is_flax_available() and is_transformers_available()):
+    if not is_transformers_flax_compatible():
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
     from ..utils import dummy_flax_and_transformers_objects  # noqa F403
@@ -930,15 +930,15 @@
             from .consisid import ConsisIDPipeline
 
         try:
-            if not is_flax_available():
+            if not is_transformers_flax_compatible():
                 raise OptionalDependencyNotAvailable()
         except OptionalDependencyNotAvailable:
             from ..utils.dummy_flax_objects import *  # noqa F403
         else:
             from .pipeline_flax_utils import FlaxDiffusionPipeline
 
         try:
-            if not (is_flax_available() and is_transformers_available()):
+            if not is_transformers_flax_compatible():
                 raise OptionalDependencyNotAvailable()
         except OptionalDependencyNotAvailable:
             from ..utils.dummy_flax_and_transformers_objects import *

src/diffusers/pipelines/controlnet/__init__.py

@@ -5,9 +5,9 @@
     OptionalDependencyNotAvailable,
     _LazyModule,
     get_objects_from_module,
-    is_flax_available,
     is_torch_available,
     is_transformers_available,
+    is_transformers_flax_compatible,
 )
 
 
@@ -34,7 +34,7 @@
     _import_structure["pipeline_controlnet_union_sd_xl"] = ["StableDiffusionXLControlNetUnionPipeline"]
     _import_structure["pipeline_controlnet_union_sd_xl_img2img"] = ["StableDiffusionXLControlNetUnionImg2ImgPipeline"]
 try:
-    if not (is_transformers_available() and is_flax_available()):
+    if not is_transformers_flax_compatible():
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
     from ...utils import dummy_flax_and_transformers_objects  # noqa F403
@@ -65,7 +65,7 @@
         from .pipeline_controlnet_union_sd_xl_img2img import StableDiffusionXLControlNetUnionImg2ImgPipeline
 
     try:
-        if not (is_transformers_available() and is_flax_available()):
+        if not is_transformers_flax_compatible():
             raise OptionalDependencyNotAvailable()
     except OptionalDependencyNotAvailable:
         from ...utils.dummy_flax_and_transformers_objects import *  # noqa F403

src/diffusers/pipelines/controlnet_sd3/__init__.py

@@ -5,9 +5,9 @@
     OptionalDependencyNotAvailable,
     _LazyModule,
     get_objects_from_module,
-    is_flax_available,
     is_torch_available,
     is_transformers_available,
+    is_transformers_flax_compatible,
 )
 
 
@@ -39,7 +39,7 @@
         from .pipeline_stable_diffusion_3_controlnet_inpainting import StableDiffusion3ControlNetInpaintingPipeline
 
     try:
-        if not (is_transformers_available() and is_flax_available()):
+        if not is_transformers_flax_compatible():
             raise OptionalDependencyNotAvailable()
     except OptionalDependencyNotAvailable:
         from ...utils.dummy_flax_and_transformers_objects import *  # noqa F403

src/diffusers/pipelines/deprecated/controlnet_xs/__init__.py

@@ -5,9 +5,9 @@
     OptionalDependencyNotAvailable,
     _LazyModule,
     get_objects_from_module,
-    is_flax_available,
     is_torch_available,
     is_transformers_available,
+    is_transformers_flax_compatible,
 )
 
 
@@ -25,7 +25,7 @@
     _import_structure["pipeline_controlnet_xs"] = ["StableDiffusionControlNetXSPipeline"]
     _import_structure["pipeline_controlnet_xs_sd_xl"] = ["StableDiffusionXLControlNetXSPipeline"]
 try:
-    if not (is_transformers_available() and is_flax_available()):
+    if not is_transformers_flax_compatible():
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
     from ....utils import dummy_flax_and_transformers_objects  # noqa F403
@@ -47,7 +47,7 @@
         from .pipeline_controlnet_xs_sd_xl import StableDiffusionXLControlNetXSPipeline
 
     try:
-        if not (is_transformers_available() and is_flax_available()):
+        if not is_transformers_flax_compatible():
             raise OptionalDependencyNotAvailable()
     except OptionalDependencyNotAvailable:
         from ....utils.dummy_flax_and_transformers_objects import *  # noqa F403

src/diffusers/pipelines/ernie_image/pipeline_ernie_image.py

@@ -20,18 +20,25 @@
 from typing import Callable, List, Optional, Union
 
 import torch
-from transformers import AutoModel, AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoTokenizer, Mistral3Model
 
 from ...image_processor import VaeImageProcessor
 from ...loaders import ErnieImageLoraLoaderMixin
 from ...models import AutoencoderKLFlux2
 from ...models.transformers import ErnieImageTransformer2DModel
 from ...pipelines.pipeline_utils import DiffusionPipeline
 from ...schedulers import FlowMatchEulerDiscreteScheduler
+from ...utils.import_utils import is_transformers_version
 from ...utils.torch_utils import randn_tensor
 from .pipeline_output import ErnieImagePipelineOutput
 
 
+if is_transformers_version("<", "5.0.0"):
+    raise ImportError("`ErnieImagePipeline` requires `transformers>=5.0.0` for `Ministral3ForCausalLM`.")
+
+from transformers import Ministral3ForCausalLM  # noqa: E402
+
+
 class ErnieImagePipeline(DiffusionPipeline, ErnieImageLoraLoaderMixin):
     """
     Pipeline for text-to-image generation using ErnieImageTransformer2DModel.
@@ -52,10 +59,10 @@ def __init__(
         self,
         transformer: ErnieImageTransformer2DModel,
         vae: AutoencoderKLFlux2,
-        text_encoder: AutoModel,
+        text_encoder: Mistral3Model,
         tokenizer: AutoTokenizer,
         scheduler: FlowMatchEulerDiscreteScheduler,
-        pe: Optional[AutoModelForCausalLM] = None,
+        pe: Optional[Ministral3ForCausalLM] = None,
         pe_tokenizer: Optional[AutoTokenizer] = None,
     ):
         super().__init__()