Merge branch 'main' into sf-modular

docs/source/en/_toctree.yml

@@ -529,6 +529,8 @@
         title: Kandinsky 2.2
       - local: api/pipelines/kandinsky3
         title: Kandinsky 3
+      - local: api/pipelines/kandinsky5
+        title: Kandinsky 5
       - local: api/pipelines/kolors
         title: Kolors
       - local: api/pipelines/latent_consistency_models
@@ -636,8 +638,6 @@
         title: HunyuanVideo
       - local: api/pipelines/i2vgenxl
         title: I2VGen-XL
-      - local: api/pipelines/kandinsky5_video
-        title: Kandinsky 5.0 Video
       - local: api/pipelines/latte
         title: Latte
       - local: api/pipelines/ltx_video

docs/source/en/api/pipelines/kandinsky5.md

@@ -7,9 +7,9 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# Kandinsky 5.0 Video
+# Kandinsky 5.0
 
-Kandinsky 5.0 Video is created by the Kandinsky team: Alexey Letunovskiy, Maria Kovaleva, Ivan Kirillov, Lev Novitskiy, Denis Koposov, Dmitrii Mikhailov, Anna Averchenkova, Andrey Shutkin, Julia Agafonova, Olga Kim, Anastasiia Kargapoltseva, Nikita Kiselev, Anna Dmitrienko,  Anastasia Maltseva, Kirill Chernyshev, Ilia Vasiliev, Viacheslav Vasilev, Vladimir Polovnikov, Yury Kolabushin, Alexander Belykh, Mikhail Mamaev, Anastasia Aliaskina, Tatiana Nikulina, Polina Gavrilova, Vladimir Arkhipkin, Vladimir Korviakov, Nikolai Gerasimenko, Denis Parkhomenko, Denis Dimitrov
+Kandinsky 5.0 is created by the Kandinsky team: Alexey Letunovskiy, Maria Kovaleva, Ivan Kirillov, Lev Novitskiy, Denis Koposov, Dmitrii Mikhailov, Anna Averchenkova, Andrey Shutkin, Julia Agafonova, Olga Kim, Anastasiia Kargapoltseva, Nikita Kiselev, Anna Dmitrienko,  Anastasia Maltseva, Kirill Chernyshev, Ilia Vasiliev, Viacheslav Vasilev, Vladimir Polovnikov, Yury Kolabushin, Alexander Belykh, Mikhail Mamaev, Anastasia Aliaskina, Tatiana Nikulina, Polina Gavrilova, Vladimir Arkhipkin, Vladimir Korviakov, Nikolai Gerasimenko, Denis Parkhomenko, Denis Dimitrov
 
 
 Kandinsky 5.0 is a family of diffusion models for Video & Image generation. Kandinsky 5.0 T2V Lite is a lightweight video generation model (2B parameters) that ranks #1 among open-source models in its class. It outperforms larger models and offers the best understanding of Russian concepts in the open-source ecosystem.
@@ -92,7 +92,7 @@ pipe = pipe.to("cuda")
 
 pipe.transformer.set_attention_backend(
     "flex"
-)                                       # <--- Sett attention bakend to Flex
+)                                       # <--- Set attention backend to Flex
 pipe.transformer.compile(
     mode="max-autotune-no-cudagraphs", 
     dynamic=True
@@ -115,7 +115,7 @@ export_to_video(output, "output.mp4", fps=24, quality=9)
 ```
 
 ### Diffusion Distilled model
-**⚠️ Warning!** all nocfg and diffusion distilled models should be infered wothout CFG (```guidance_scale=1.0```):
+**⚠️ Warning!** all nocfg and diffusion distilled models should be inferred without CFG (```guidance_scale=1.0```):
 
 ```python
 model_id = "ai-forever/Kandinsky-5.0-T2V-Lite-distilled16steps-5s-Diffusers"

docs/source/en/modular_diffusers/guiders.md

@@ -159,7 +159,7 @@ Change the [`~ComponentSpec.default_creation_method`] to `from_pretrained` and u
 ```py
 guider_spec = t2i_pipeline.get_component_spec("guider")
 guider_spec.default_creation_method="from_pretrained"
-guider_spec.repo="YiYiXu/modular-loader-t2i-guider"
+guider_spec.pretrained_model_name_or_path="YiYiXu/modular-loader-t2i-guider"
 guider_spec.subfolder="pag_guider"
 pag_guider = guider_spec.load()
 t2i_pipeline.update_components(guider=pag_guider)

docs/source/en/modular_diffusers/modular_pipeline.md

@@ -313,14 +313,14 @@ unet_spec
 ComponentSpec(
     name='unet',
     type_hint=<class 'diffusers.models.unets.unet_2d_condition.UNet2DConditionModel'>,
-    repo='RunDiffusion/Juggernaut-XL-v9',
+    pretrained_model_name_or_path='RunDiffusion/Juggernaut-XL-v9',
     subfolder='unet',
     variant='fp16',
     default_creation_method='from_pretrained'
 )
 
 # modify to load from a different repository
-unet_spec.repo = "stabilityai/stable-diffusion-xl-base-1.0"
+unet_spec.pretrained_model_name_or_path = "stabilityai/stable-diffusion-xl-base-1.0"
 
 # load component with modified spec
 unet = unet_spec.load(torch_dtype=torch.float16)

docs/source/zh/modular_diffusers/guiders.md

@@ -157,7 +157,7 @@ guider.push_to_hub("YiYiXu/modular-loader-t2i-guider", subfolder="pag_guider")
 ```py
 guider_spec = t2i_pipeline.get_component_spec("guider")
 guider_spec.default_creation_method="from_pretrained"
-guider_spec.repo="YiYiXu/modular-loader-t2i-guider"
+guider_spec.pretrained_model_name_or_path="YiYiXu/modular-loader-t2i-guider"
 guider_spec.subfolder="pag_guider"
 pag_guider = guider_spec.load()
 t2i_pipeline.update_components(guider=pag_guider)

docs/source/zh/modular_diffusers/modular_pipeline.md

@@ -313,14 +313,14 @@ unet_spec
 ComponentSpec(
     name='unet',
     type_hint=<class 'diffusers.models.unets.unet_2d_condition.UNet2DConditionModel'>,
-    repo='RunDiffusion/Juggernaut-XL-v9',
+    pretrained_model_name_or_path='RunDiffusion/Juggernaut-XL-v9',
     subfolder='unet',
     variant='fp16',
     default_creation_method='from_pretrained'
 )
 
 # 修改以从不同的仓库加载
-unet_spec.repo = "stabilityai/stable-diffusion-xl-base-1.0"
+unet_spec.pretrained_model_name_or_path = "stabilityai/stable-diffusion-xl-base-1.0"
 
 # 使用修改后的规范加载组件
 unet = unet_spec.load(torch_dtype=torch.float16)

src/diffusers/loaders/lora_conversion_utils.py

@@ -2213,10 +2213,6 @@ def _convert_non_diffusers_qwen_lora_to_diffusers(state_dict):
 
         state_dict = {convert_key(k): v for k, v in state_dict.items()}
 
-    has_default = any("default." in k for k in state_dict)
-    if has_default:
-        state_dict = {k.replace("default.", ""): v for k, v in state_dict.items()}
-
     converted_state_dict = {}
     all_keys = list(state_dict.keys())
     down_key = ".lora_down.weight"

src/diffusers/loaders/lora_pipeline.py

@@ -4940,8 +4940,7 @@ class QwenImageLoraLoaderMixin(LoraBaseMixin):
         has_alphas_in_sd = any(k.endswith(".alpha") for k in state_dict)
         has_lora_unet = any(k.startswith("lora_unet_") for k in state_dict)
         has_diffusion_model = any(k.startswith("diffusion_model.") for k in state_dict)
-        has_default = any("default." in k for k in state_dict)
-        if has_alphas_in_sd or has_lora_unet or has_diffusion_model or has_default:
+        if has_alphas_in_sd or has_lora_unet or has_diffusion_model:
             state_dict = _convert_non_diffusers_qwen_lora_to_diffusers(state_dict)
 
         out = (state_dict, metadata) if return_lora_metadata else state_dict

src/diffusers/loaders/single_file_utils.py

@@ -387,6 +387,14 @@ def is_valid_url(url):
     return False
 
 
+def _is_single_file_path_or_url(pretrained_model_name_or_path):
+    if not os.path.isfile(pretrained_model_name_or_path) or not is_valid_url(pretrained_model_name_or_path):
+        return False
+
+    repo_id, weight_name = _extract_repo_id_and_weights_name(pretrained_model_name_or_path)
+    return bool(repo_id and weight_name)
+
+
 def _extract_repo_id_and_weights_name(pretrained_model_name_or_path):
     if not is_valid_url(pretrained_model_name_or_path):
         raise ValueError("Invalid `pretrained_model_name_or_path` provided. Please set it to a valid URL.")
@@ -398,7 +406,6 @@ def _extract_repo_id_and_weights_name(pretrained_model_name_or_path):
         pretrained_model_name_or_path = pretrained_model_name_or_path.replace(prefix, "")
     match = re.match(pattern, pretrained_model_name_or_path)
     if not match:
-        logger.warning("Unable to identify the repo_id and weights_name from the provided URL.")
         return repo_id, weights_name
 
     repo_id = f"{match.group(1)}/{match.group(2)}"

src/diffusers/models/attention_dispatch.py

@@ -649,86 +649,6 @@ def _(
 # ===== Helper functions to use attention backends with templated CP autograd functions =====
 
 
-def _native_attention_forward_op(
-    ctx: torch.autograd.function.FunctionCtx,
-    query: torch.Tensor,
-    key: torch.Tensor,
-    value: torch.Tensor,
-    attn_mask: Optional[torch.Tensor] = None,
-    dropout_p: float = 0.0,
-    is_causal: bool = False,
-    scale: Optional[float] = None,
-    enable_gqa: bool = False,
-    return_lse: bool = False,
-    _save_ctx: bool = True,
-    _parallel_config: Optional["ParallelConfig"] = None,
-):
-    # Native attention does not return_lse
-    if return_lse:
-        raise ValueError("Native attention does not support return_lse=True")
-
-    # used for backward pass
-    if _save_ctx:
-        ctx.save_for_backward(query, key, value)
-        ctx.attn_mask = attn_mask
-        ctx.dropout_p = dropout_p
-        ctx.is_causal = is_causal
-        ctx.scale = scale
-        ctx.enable_gqa = enable_gqa
-
-    query, key, value = (x.permute(0, 2, 1, 3) for x in (query, key, value))
-    out = torch.nn.functional.scaled_dot_product_attention(
-        query=query,
-        key=key,
-        value=value,
-        attn_mask=attn_mask,
-        dropout_p=dropout_p,
-        is_causal=is_causal,
-        scale=scale,
-        enable_gqa=enable_gqa,
-    )
-    out = out.permute(0, 2, 1, 3)
-
-    return out
-
-
-def _native_attention_backward_op(
-    ctx: torch.autograd.function.FunctionCtx,
-    grad_out: torch.Tensor,
-    *args,
-    **kwargs,
-):
-    query, key, value = ctx.saved_tensors
-
-    query.requires_grad_(True)
-    key.requires_grad_(True)
-    value.requires_grad_(True)
-
-    query_t, key_t, value_t = (x.permute(0, 2, 1, 3) for x in (query, key, value))
-    out = torch.nn.functional.scaled_dot_product_attention(
-        query=query_t,
-        key=key_t,
-        value=value_t,
-        attn_mask=ctx.attn_mask,
-        dropout_p=ctx.dropout_p,
-        is_causal=ctx.is_causal,
-        scale=ctx.scale,
-        enable_gqa=ctx.enable_gqa,
-    )
-    out = out.permute(0, 2, 1, 3)
-
-    grad_out_t = grad_out.permute(0, 2, 1, 3)
-    grad_query_t, grad_key_t, grad_value_t = torch.autograd.grad(
-        outputs=out, inputs=[query_t, key_t, value_t], grad_outputs=grad_out_t, retain_graph=False
-    )
-
-    grad_query = grad_query_t.permute(0, 2, 1, 3)
-    grad_key = grad_key_t.permute(0, 2, 1, 3)
-    grad_value = grad_value_t.permute(0, 2, 1, 3)
-
-    return grad_query, grad_key, grad_value
-
-
 # https://github.com/pytorch/pytorch/blob/8904ba638726f8c9a5aff5977c4aa76c9d2edfa6/aten/src/ATen/native/native_functions.yaml#L14958
 # forward declaration:
 #   aten::_scaled_dot_product_cudnn_attention(Tensor query, Tensor key, Tensor value, Tensor? attn_bias, bool compute_log_sumexp, float dropout_p=0., bool is_causal=False, bool return_debug_mask=False, *, float? scale=None) -> (Tensor output, Tensor logsumexp, Tensor cum_seq_q, Tensor cum_seq_k, SymInt max_q, SymInt max_k, Tensor philox_seed, Tensor philox_offset, Tensor debug_attn_mask)
@@ -1603,7 +1523,6 @@ def _native_flex_attention(
 @_AttentionBackendRegistry.register(
     AttentionBackendName.NATIVE,
     constraints=[_check_device, _check_shape],
-    supports_context_parallel=True,
 )
 def _native_attention(
     query: torch.Tensor,
@@ -1619,35 +1538,18 @@ def _native_attention(
 ) -> torch.Tensor:
     if return_lse:
         raise ValueError("Native attention backend does not support setting `return_lse=True`.")
-    if _parallel_config is None:
-        query, key, value = (x.permute(0, 2, 1, 3) for x in (query, key, value))
-        out = torch.nn.functional.scaled_dot_product_attention(
-            query=query,
-            key=key,
-            value=value,
-            attn_mask=attn_mask,
-            dropout_p=dropout_p,
-            is_causal=is_causal,
-            scale=scale,
-            enable_gqa=enable_gqa,
-        )
-        out = out.permute(0, 2, 1, 3)
-    else:
-        out = _templated_context_parallel_attention(
-            query,
-            key,
-            value,
-            attn_mask,
-            dropout_p,
-            is_causal,
-            scale,
-            enable_gqa,
-            return_lse,
-            forward_op=_native_attention_forward_op,
-            backward_op=_native_attention_backward_op,
-            _parallel_config=_parallel_config,
-        )
-
+    query, key, value = (x.permute(0, 2, 1, 3) for x in (query, key, value))
+    out = torch.nn.functional.scaled_dot_product_attention(
+        query=query,
+        key=key,
+        value=value,
+        attn_mask=attn_mask,
+        dropout_p=dropout_p,
+        is_causal=is_causal,
+        scale=scale,
+        enable_gqa=enable_gqa,
+    )
+    out = out.permute(0, 2, 1, 3)
     return out
 
 

src/diffusers/models/auto_model.py

@@ -147,13 +147,14 @@ class AutoModel(ConfigMixin):
             "force_download",
             "local_files_only",
             "proxies",
+            "resume_download",
             "revision",
             "token",
         ]
         hub_kwargs = {name: kwargs.pop(name, None) for name in hub_kwargs_names}
 
         # load_config_kwargs uses the same hub kwargs minus subfolder and resume_download
-        load_config_kwargs = {k: v for k, v in hub_kwargs.items() if k not in ["subfolder"]}
+        load_config_kwargs = {k: v for k, v in hub_kwargs.items() if k not in ["subfolder", "resume_download"]}
 
         library = None
         orig_class_name = None
@@ -204,6 +205,7 @@ class AutoModel(ConfigMixin):
                 module_file=module_file,
                 class_name=class_name,
                 **hub_kwargs,
+                **kwargs,
             )
         else:
             from ..pipelines.pipeline_loading_utils import ALL_IMPORTABLE_CLASSES, get_class_obj_and_candidates

src/diffusers/models/autoencoders/vae.py

@@ -286,9 +286,11 @@ class Decoder(nn.Module):
 
         sample = self.conv_in(sample)
 
+        upscale_dtype = next(iter(self.up_blocks.parameters())).dtype
         if torch.is_grad_enabled() and self.gradient_checkpointing:
             # middle
             sample = self._gradient_checkpointing_func(self.mid_block, sample, latent_embeds)
+            sample = sample.to(upscale_dtype)
 
             # up
             for up_block in self.up_blocks:
@@ -296,6 +298,7 @@ class Decoder(nn.Module):
         else:
             # middle
             sample = self.mid_block(sample, latent_embeds)
+            sample = sample.to(upscale_dtype)
 
             # up
             for up_block in self.up_blocks:

src/diffusers/modular_pipelines/components_manager.py

@@ -164,11 +164,7 @@ class AutoOffloadStrategy:
 
         device_type = execution_device.type
         device_module = getattr(torch, device_type, torch.cuda)
-        try:
-            mem_on_device = device_module.mem_get_info(execution_device.index)[0]
-        except AttributeError:
-            raise AttributeError(f"Do not know how to obtain obtain memory info for {str(device_module)}.")
-
+        mem_on_device = device_module.mem_get_info(execution_device.index)[0]
         mem_on_device = mem_on_device - self.memory_reserve_margin
         if current_module_size < mem_on_device:
             return []
@@ -703,8 +699,6 @@ class ComponentsManager:
         if not is_accelerate_available():
             raise ImportError("Make sure to install accelerate to use auto_cpu_offload")
 
-        # TODO: add a warning if mem_get_info isn't available on `device`.
-
         for name, component in self.components.items():
             if isinstance(component, torch.nn.Module) and hasattr(component, "_hf_hook"):
                 remove_hook_from_module(component, recurse=True)

src/diffusers/modular_pipelines/flux/before_denoise.py

@@ -598,7 +598,7 @@ class FluxKontextRoPEInputsStep(ModularPipelineBlocks):
             and getattr(block_state, "image_width", None) is not None
         ):
             image_latent_height = 2 * (int(block_state.image_height) // (components.vae_scale_factor * 2))
-            image_latent_width = 2 * (int(block_state.image_width) // (components.vae_scale_factor * 2))
+            image_latent_width = 2 * (int(block_state.width) // (components.vae_scale_factor * 2))
             img_ids = FluxPipeline._prepare_latent_image_ids(
                 None, image_latent_height // 2, image_latent_width // 2, device, dtype
             )

src/diffusers/modular_pipelines/flux/denoise.py

@@ -59,7 +59,7 @@ class FluxLoopDenoiser(ModularPipelineBlocks):
             ),
             InputParam(
                 "guidance",
-                required=False,
+                required=True,
                 type_hint=torch.Tensor,
                 description="Guidance scale as a tensor",
             ),
@@ -141,7 +141,7 @@ class FluxKontextLoopDenoiser(ModularPipelineBlocks):
             ),
             InputParam(
                 "guidance",
-                required=False,
+                required=True,
                 type_hint=torch.Tensor,
                 description="Guidance scale as a tensor",
             ),

src/diffusers/modular_pipelines/flux/encoders.py

@@ -95,7 +95,7 @@ class FluxProcessImagesInputStep(ModularPipelineBlocks):
             ComponentSpec(
                 "image_processor",
                 VaeImageProcessor,
-                config=FrozenDict({"vae_scale_factor": 16, "vae_latent_channels": 16}),
+                config=FrozenDict({"vae_scale_factor": 16}),
                 default_creation_method="from_config",
             ),
         ]
@@ -143,6 +143,10 @@ class FluxProcessImagesInputStep(ModularPipelineBlocks):
 class FluxKontextProcessImagesInputStep(ModularPipelineBlocks):
     model_name = "flux-kontext"
 
+    def __init__(self, _auto_resize=True):
+        self._auto_resize = _auto_resize
+        super().__init__()
+
     @property
     def description(self) -> str:
         return (
@@ -163,7 +167,7 @@ class FluxKontextProcessImagesInputStep(ModularPipelineBlocks):
 
     @property
     def inputs(self) -> List[InputParam]:
-        return [InputParam("image"), InputParam("_auto_resize", type_hint=bool, default=True)]
+        return [InputParam("image")]
 
     @property
     def intermediate_outputs(self) -> List[OutputParam]:
@@ -191,8 +195,7 @@ class FluxKontextProcessImagesInputStep(ModularPipelineBlocks):
             img = images[0]
             image_height, image_width = components.image_processor.get_default_height_width(img)
             aspect_ratio = image_width / image_height
-            _auto_resize = block_state._auto_resize
-            if _auto_resize:
+            if self._auto_resize:
                 # Kontext is trained on specific resolutions, using one of them is recommended
                 _, image_width, image_height = min(
                     (abs(aspect_ratio - w / h), w, h) for w, h in PREFERRED_KONTEXT_RESOLUTIONS

src/diffusers/modular_pipelines/flux/inputs.py

@@ -112,10 +112,6 @@ class FluxTextInputStep(ModularPipelineBlocks):
         block_state.prompt_embeds = block_state.prompt_embeds.view(
             block_state.batch_size * block_state.num_images_per_prompt, seq_len, -1
         )
-        pooled_prompt_embeds = block_state.pooled_prompt_embeds.repeat(1, block_state.num_images_per_prompt)
-        block_state.pooled_prompt_embeds = pooled_prompt_embeds.view(
-            block_state.batch_size * block_state.num_images_per_prompt, -1
-        )
         self.set_block_state(state, block_state)
 
         return components, state

src/diffusers/modular_pipelines/modular_pipeline.py

@@ -305,15 +305,15 @@ class ModularPipelineBlocks(ConfigMixin, PushToHubMixin):
             "cache_dir",
             "force_download",
             "local_files_only",
-            "local_dir",
             "proxies",
+            "resume_download",
             "revision",
             "subfolder",
             "token",
         ]
         hub_kwargs = {name: kwargs.pop(name) for name in hub_kwargs_names if name in kwargs}
 
-        config = cls.load_config(pretrained_model_name_or_path, **hub_kwargs)
+        config = cls.load_config(pretrained_model_name_or_path)
         has_remote_code = "auto_map" in config and cls.__name__ in config["auto_map"]
         trust_remote_code = resolve_trust_remote_code(
             trust_remote_code, pretrained_model_name_or_path, has_remote_code
@@ -331,10 +331,11 @@ class ModularPipelineBlocks(ConfigMixin, PushToHubMixin):
             module_file=module_file,
             class_name=class_name,
             **hub_kwargs,
+            **kwargs,
         )
         expected_kwargs, optional_kwargs = block_cls._get_signature_keys(block_cls)
         block_kwargs = {
-            name: kwargs.get(name) for name in kwargs if name in expected_kwargs or name in optional_kwargs
+            name: kwargs.pop(name) for name in kwargs if name in expected_kwargs or name in optional_kwargs
         }
 
         return block_cls(**block_kwargs)
@@ -360,7 +361,7 @@ class ModularPipelineBlocks(ConfigMixin, PushToHubMixin):
         collection: Optional[str] = None,
     ) -> "ModularPipeline":
         """
-        create a ModularPipeline, optionally accept modular_repo to load from hub.
+        create a ModularPipeline, optionally accept pretrained_model_name_or_path to load from hub.
         """
         pipeline_class_name = MODULAR_PIPELINE_MAPPING.get(self.model_name, ModularPipeline.__name__)
         diffusers_module = importlib.import_module("diffusers")
@@ -1561,7 +1562,7 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
                         if name in self._component_specs and isinstance(value, (tuple, list)) and len(value) == 2:
                             library, class_name = value
                             component_spec_dict = {
-                                "repo": pretrained_model_name_or_path,
+                                "pretrained_model_name_or_path": pretrained_model_name_or_path,
                                 "subfolder": name,
                                 "type_hint": (library, class_name),
                             }
@@ -1618,8 +1619,8 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
             pretrained_model_name_or_path (`str` or `os.PathLike`, optional):
                 Path to a pretrained pipeline configuration. It will first try to load config from
                 `modular_model_index.json`, then fallback to `model_index.json` for compatibility with standard
-                non-modular repositories. If the repo does not contain any pipeline config, it will be set to None
-                during initialization.
+                non-modular repositories. If the pretrained_model_name_or_path does not contain any pipeline config, it
+                will be set to None during initialization.
             trust_remote_code (`bool`, optional):
                 Whether to trust remote code when loading the pipeline, need to be set to True if you want to create
                 pipeline blocks based on the custom code in `pretrained_model_name_or_path`
@@ -1789,7 +1790,7 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
                 library, class_name = None, None
 
             # extract the loading spec from the updated component spec that'll be used as part of modular_model_index.json config
-            # e.g. {"repo": "stabilityai/stable-diffusion-2-1",
+            # e.g. {"pretrained_model_name_or_path": "stabilityai/stable-diffusion-2-1",
             #       "type_hint": ("diffusers", "UNet2DConditionModel"),
             #       "subfolder": "unet",
             #       "variant": None,
@@ -2093,8 +2094,10 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
             **kwargs: additional kwargs to be passed to `from_pretrained()`.Can be:
              - a single value to be applied to all components to be loaded, e.g. torch_dtype=torch.bfloat16
              - a dict, e.g. torch_dtype={"unet": torch.bfloat16, "default": torch.float32}
-             - if potentially override ComponentSpec if passed a different loading field in kwargs, e.g. `repo`,
-               `variant`, `revision`, etc.
+             - if potentially override ComponentSpec if passed a different loading field in kwargs, e.g.
+               `pretrained_model_name_or_path`, `variant`, `revision`, etc.
+             - if potentially override ComponentSpec if passed a different loading field in kwargs, e.g.
+               `pretrained_model_name_or_path`, `variant`, `revision`, etc.
         """
 
         if names is None:
@@ -2130,13 +2133,8 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
                         component_load_kwargs[key] = value["default"]
             try:
                 components_to_register[name] = spec.load(**component_load_kwargs)
-            except Exception:
-                logger.warning(
-                    f"\nFailed to create component {name}:\n"
-                    f"- Component spec: {spec}\n"
-                    f"- load() called with kwargs: {component_load_kwargs}\n\n"
-                    f"{traceback.format_exc()}"
-                )
+            except Exception as e:
+                logger.warning(f"Failed to create component '{name}': {e}")
 
         # Register all components at once
         self.register_components(**components_to_register)
@@ -2358,10 +2356,10 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
           - "type_hint": Tuple[str, str]
               Library name and class name of the component. (e.g. ("diffusers", "UNet2DConditionModel"))
           - All loading fields defined by `component_spec.loading_fields()`, typically:
-              - "repo": Optional[str]
-                  The model repository (e.g., "stabilityai/stable-diffusion-xl").
+              - "pretrained_model_name_or_path": Optional[str]
+                  The model pretrained_model_name_or_pathsitory (e.g., "stabilityai/stable-diffusion-xl").
               - "subfolder": Optional[str]
-                  A subfolder within the repo where this component lives.
+                  A subfolder within the pretrained_model_name_or_path where this component lives.
               - "variant": Optional[str]
                   An optional variant identifier for the model.
               - "revision": Optional[str]
@@ -2378,12 +2376,14 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
         Example:
             >>> from diffusers.pipelines.modular_pipeline_utils import ComponentSpec >>> from diffusers import
             UNet2DConditionModel >>> spec = ComponentSpec(
-                ... name="unet", ... type_hint=UNet2DConditionModel, ... config=None, ... repo="path/to/repo", ...
-                subfolder="subfolder", ... variant=None, ... revision=None, ...
-                default_creation_method="from_pretrained",
+                ... name="unet", ... type_hint=UNet2DConditionModel, ... config=None, ...
+                pretrained_model_name_or_path="path/to/pretrained_model_name_or_path", ... subfolder="subfolder", ...
+                variant=None, ... revision=None, ... default_creation_method="from_pretrained",
             ... ) >>> ModularPipeline._component_spec_to_dict(spec) {
-                "type_hint": ("diffusers", "UNet2DConditionModel"), "repo": "path/to/repo", "subfolder": "subfolder",
-                "variant": None, "revision": None,
+                "type_hint": ("diffusers", "UNet2DConditionModel"), "pretrained_model_name_or_path": "path/to/repo",
+                "subfolder": "subfolder", "variant": None, "revision": None,
+                "type_hint": ("diffusers", "UNet2DConditionModel"), "pretrained_model_name_or_path": "path/to/repo",
+                "subfolder": "subfolder", "variant": None, "revision": None,
             }
         """
         if component_spec.default_creation_method != "from_pretrained":
@@ -2412,10 +2412,10 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
           - "type_hint": Tuple[str, str]
               Library name and class name of the component. (e.g. ("diffusers", "UNet2DConditionModel"))
           - All loading fields defined by `component_spec.loading_fields()`, typically:
-              - "repo": Optional[str]
+              - "pretrained_model_name_or_path": Optional[str]
                   The model repository (e.g., "stabilityai/stable-diffusion-xl").
               - "subfolder": Optional[str]
-                  A subfolder within the repo where this component lives.
+                  A subfolder within the pretrained_model_name_or_path where this component lives.
               - "variant": Optional[str]
                   An optional variant identifier for the model.
               - "revision": Optional[str]
@@ -2432,11 +2432,20 @@ class ModularPipeline(ConfigMixin, PushToHubMixin):
             ComponentSpec: A reconstructed ComponentSpec object.
 
         Example:
-            >>> spec_dict = { ... "type_hint": ("diffusers", "UNet2DConditionModel"), ... "repo":
-            "stabilityai/stable-diffusion-xl", ... "subfolder": "unet", ... "variant": None, ... "revision": None, ...
-            } >>> ModularPipeline._dict_to_component_spec("unet", spec_dict) ComponentSpec(
-                name="unet", type_hint=UNet2DConditionModel, config=None, repo="stabilityai/stable-diffusion-xl",
-                subfolder="unet", variant=None, revision=None, default_creation_method="from_pretrained"
+            >>> spec_dict = { ... "type_hint": ("diffusers", "UNet2DConditionModel"), ...
+            "pretrained_model_name_or_path": "stabilityai/stable-diffusion-xl", ... "subfolder": "unet", ... "variant":
+            None, ... "revision": None, ... } >>> ModularPipeline._dict_to_component_spec("unet", spec_dict)
+            ComponentSpec(
+                name="unet", type_hint=UNet2DConditionModel, config=None,
+                pretrained_model_name_or_path="stabilityai/stable-diffusion-xl", subfolder="unet", variant=None,
+                revision=None, default_creation_method="from_pretrained"
+            >>> spec_dict = { ... "type_hint": ("diffusers", "UNet2DConditionModel"), ...
+            "pretrained_model_name_or_path": "stabilityai/stable-diffusion-xl", ... "subfolder": "unet", ... "variant":
+            None, ... "revision": None, ... } >>> ModularPipeline._dict_to_component_spec("unet", spec_dict)
+            ComponentSpec(
+                name="unet", type_hint=UNet2DConditionModel, config=None,
+                pretrained_model_name_or_path="stabilityai/stable-diffusion-xl", subfolder="unet", variant=None,
+                revision=None, default_creation_method="from_pretrained"
             )
         """
         # make a shallow copy so we can pop() safely

src/diffusers/modular_pipelines/modular_pipeline_utils.py

@@ -21,6 +21,7 @@ from typing import Any, Dict, List, Literal, Optional, Type, Union
 import torch
 
 from ..configuration_utils import ConfigMixin, FrozenDict
+from ..loaders.single_file_utils import _is_single_file_path_or_url
 from ..utils import is_torch_available, logging
 
 
@@ -80,10 +81,10 @@ class ComponentSpec:
         type_hint: Type of the component (e.g. UNet2DConditionModel)
         description: Optional description of the component
         config: Optional config dict for __init__ creation
-        repo: Optional repo path for from_pretrained creation
-        subfolder: Optional subfolder in repo
-        variant: Optional variant in repo
-        revision: Optional revision in repo
+        pretrained_model_name_or_path: Optional pretrained_model_name_or_path path for from_pretrained creation
+        subfolder: Optional subfolder in pretrained_model_name_or_path
+        variant: Optional variant in pretrained_model_name_or_path
+        revision: Optional revision in pretrained_model_name_or_path
         default_creation_method: Preferred creation method - "from_config" or "from_pretrained"
     """
 
@@ -91,13 +92,20 @@ class ComponentSpec:
     type_hint: Optional[Type] = None
     description: Optional[str] = None
     config: Optional[FrozenDict] = None
-    # YiYi Notes: should we change it to pretrained_model_name_or_path for consistency? a bit long for a field name
-    repo: Optional[Union[str, List[str]]] = field(default=None, metadata={"loading": True})
+    pretrained_model_name_or_path: Optional[Union[str, List[str]]] = field(default=None, metadata={"loading": True})
     subfolder: Optional[str] = field(default="", metadata={"loading": True})
     variant: Optional[str] = field(default=None, metadata={"loading": True})
     revision: Optional[str] = field(default=None, metadata={"loading": True})
     default_creation_method: Literal["from_config", "from_pretrained"] = "from_pretrained"
 
+    # Deprecated
+    repo: Optional[Union[str, List[str]]] = field(default=None, metadata={"loading": False})
+
+    def __post_init__(self):
+        repo_value = self.repo
+        if repo_value is not None and self.pretrained_model_name_or_path is None:
+            object.__setattr__(self, "pretrained_model_name_or_path", repo_value)
+
     def __hash__(self):
         """Make ComponentSpec hashable, using load_id as the hash value."""
         return hash((self.name, self.load_id, self.default_creation_method))
@@ -182,8 +190,8 @@ class ComponentSpec:
     @property
     def load_id(self) -> str:
         """
-        Unique identifier for this spec's pretrained load, composed of repo|subfolder|variant|revision (no empty
-        segments).
+        Unique identifier for this spec's pretrained load, composed of
+        pretrained_model_name_or_path|subfolder|variant|revision (no empty segments).
         """
         if self.default_creation_method == "from_config":
             return "null"
@@ -197,12 +205,13 @@ class ComponentSpec:
         Decode a load_id string back into a dictionary of loading fields and values.
 
         Args:
-            load_id: The load_id string to decode, format: "repo|subfolder|variant|revision"
+            load_id: The load_id string to decode, format: "pretrained_model_name_or_path|subfolder|variant|revision"
                      where None values are represented as "null"
 
         Returns:
             Dict mapping loading field names to their values. e.g. {
-                "repo": "path/to/repo", "subfolder": "subfolder", "variant": "variant", "revision": "revision"
+                "pretrained_model_name_or_path": "path/to/repo", "subfolder": "subfolder", "variant": "variant",
+                "revision": "revision"
             } If a segment value is "null", it's replaced with None. Returns None if load_id is "null" (indicating
             component not created with `load` method).
         """
@@ -259,34 +268,45 @@ class ComponentSpec:
     # YiYi TODO: add guard for type of model, if it is supported by from_pretrained
     def load(self, **kwargs) -> Any:
         """Load component using from_pretrained."""
-
-        # select loading fields from kwargs passed from user: e.g. repo, subfolder, variant, revision, note the list could change
+        # select loading fields from kwargs passed from user: e.g. pretrained_model_name_or_path, subfolder, variant, revision, note the list could change
         passed_loading_kwargs = {key: kwargs.pop(key) for key in self.loading_fields() if key in kwargs}
         # merge loading field value in the spec with user passed values to create load_kwargs
         load_kwargs = {key: passed_loading_kwargs.get(key, getattr(self, key)) for key in self.loading_fields()}
-        # repo is a required argument for from_pretrained, a.k.a. pretrained_model_name_or_path
-        repo = load_kwargs.pop("repo", None)
-        if repo is None:
+
+        pretrained_model_name_or_path = load_kwargs.pop("pretrained_model_name_or_path", None)
+        if pretrained_model_name_or_path is None:
             raise ValueError(
-                "`repo` info is required when using `load` method (you can directly set it in `repo` field of the ComponentSpec or pass it as an argument)"
+                "`pretrained_model_name_or_path` info is required when using `load` method (you can directly set it in `pretrained_model_name_or_path` field of the ComponentSpec or pass it as an argument)"
+            )
+        is_single_file = _is_single_file_path_or_url(pretrained_model_name_or_path)
+        if is_single_file and self.type_hint is None:
+            raise ValueError(
+                f"`type_hint` is required when loading a single file model but is missing for component: {self.name}"
             )
 
         if self.type_hint is None:
             try:
                 from diffusers import AutoModel
 
-                component = AutoModel.from_pretrained(repo, **load_kwargs, **kwargs)
+                component = AutoModel.from_pretrained(pretrained_model_name_or_path, **load_kwargs, **kwargs)
             except Exception as e:
                 raise ValueError(f"Unable to load {self.name} without `type_hint`: {e}")
             # update type_hint if AutoModel load successfully
             self.type_hint = component.__class__
         else:
+            # determine load method
+            load_method = (
+                getattr(self.type_hint, "from_single_file")
+                if is_single_file
+                else getattr(self.type_hint, "from_pretrained")
+            )
+
             try:
-                component = self.type_hint.from_pretrained(repo, **load_kwargs, **kwargs)
+                component = load_method(pretrained_model_name_or_path, **load_kwargs, **kwargs)
             except Exception as e:
                 raise ValueError(f"Unable to load {self.name} using load method: {e}")
 
-        self.repo = repo
+        self.pretrained_model_name_or_path = pretrained_model_name_or_path
         for k, v in load_kwargs.items():
             setattr(self, k, v)
         component._diffusers_load_id = self.load_id

src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet.py

@@ -355,7 +355,7 @@ class StableDiffusion3ControlNetPipeline(
         prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
         prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt)
+        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt, 1)
         pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)
 
         return prompt_embeds, pooled_prompt_embeds

src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py

@@ -373,7 +373,7 @@ class StableDiffusion3ControlNetInpaintingPipeline(
         prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
         prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt)
+        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt, 1)
         pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)
 
         return prompt_embeds, pooled_prompt_embeds

src/diffusers/pipelines/pag/pipeline_pag_sd_3.py

@@ -326,7 +326,7 @@ class StableDiffusion3PAGPipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSin
         prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
         prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt)
+        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt, 1)
         pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)
 
         return prompt_embeds, pooled_prompt_embeds

src/diffusers/pipelines/pag/pipeline_pag_sd_3_img2img.py

@@ -342,7 +342,7 @@ class StableDiffusion3PAGImg2ImgPipeline(DiffusionPipeline, SD3LoraLoaderMixin,
         prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
         prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt)
+        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt, 1)
         pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)
 
         return prompt_embeds, pooled_prompt_embeds

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

@@ -336,7 +336,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
         prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt)
+        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt, 1)
         pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)
 
         return prompt_embeds, pooled_prompt_embeds

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py

@@ -361,7 +361,7 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline, SD3LoraLoaderMixin, Fro
         prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
         prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt)
+        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt, 1)
         pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)
 
         return prompt_embeds, pooled_prompt_embeds

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_inpaint.py

@@ -367,7 +367,7 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline, SD3LoraLoaderMixin, Fro
         prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
         prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt)
+        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt, 1)
         pooled_prompt_embeds = pooled_prompt_embeds.view(batch_size * num_images_per_prompt, -1)
 
         return prompt_embeds, pooled_prompt_embeds

src/diffusers/utils/dynamic_modules_utils.py

@@ -254,7 +254,6 @@ def get_cached_module_file(
     token: Optional[Union[bool, str]] = None,
     revision: Optional[str] = None,
     local_files_only: bool = False,
-    local_dir: Optional[str] = None,
 ):
     """
     Prepares Downloads a module from a local folder or a distant repo and returns its path inside the cached
@@ -333,7 +332,6 @@ def get_cached_module_file(
                 force_download=force_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                local_dir=local_dir,
             )
             submodule = "git"
             module_file = pretrained_model_name_or_path + ".py"
@@ -357,8 +355,6 @@ def get_cached_module_file(
                 force_download=force_download,
                 proxies=proxies,
                 local_files_only=local_files_only,
-                local_dir=local_dir,
-                revision=revision,
                 token=token,
             )
             submodule = os.path.join("local", "--".join(pretrained_model_name_or_path.split("/")))
@@ -419,7 +415,6 @@ def get_cached_module_file(
                     token=token,
                     revision=revision,
                     local_files_only=local_files_only,
-                    local_dir=local_dir,
                 )
     return os.path.join(full_submodule, module_file)
 
@@ -436,7 +431,7 @@ def get_class_from_dynamic_module(
     token: Optional[Union[bool, str]] = None,
     revision: Optional[str] = None,
     local_files_only: bool = False,
-    local_dir: Optional[str] = None,
+    **kwargs,
 ):
     """
     Extracts a class from a module file, present in the local folder or repository of a model.
@@ -501,6 +496,5 @@ def get_class_from_dynamic_module(
         token=token,
         revision=revision,
         local_files_only=local_files_only,
-        local_dir=local_dir,
     )
     return get_class_in_module(class_name, final_module)

tests/modular_pipelines/flux/test_modular_pipeline_flux.py

@@ -1,130 +0,0 @@
-# coding=utf-8
-# Copyright 2025 HuggingFace Inc.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import random
-import tempfile
-import unittest
-
-import numpy as np
-import PIL
-import torch
-
-from diffusers.image_processor import VaeImageProcessor
-from diffusers.modular_pipelines import (
-    FluxAutoBlocks,
-    FluxKontextAutoBlocks,
-    FluxKontextModularPipeline,
-    FluxModularPipeline,
-    ModularPipeline,
-)
-
-from ...testing_utils import floats_tensor, torch_device
-from ..test_modular_pipelines_common import ModularPipelineTesterMixin
-
-
-class FluxModularTests:
-    pipeline_class = FluxModularPipeline
-    pipeline_blocks_class = FluxAutoBlocks
-    repo = "hf-internal-testing/tiny-flux-modular"
-
-    def get_pipeline(self, components_manager=None, torch_dtype=torch.float32):
-        pipeline = self.pipeline_blocks_class().init_pipeline(self.repo, components_manager=components_manager)
-        pipeline.load_components(torch_dtype=torch_dtype)
-        return pipeline
-
-    def get_dummy_inputs(self, device, seed=0):
-        if str(device).startswith("mps"):
-            generator = torch.manual_seed(seed)
-        else:
-            generator = torch.Generator(device=device).manual_seed(seed)
-        inputs = {
-            "prompt": "A painting of a squirrel eating a burger",
-            "generator": generator,
-            "num_inference_steps": 2,
-            "guidance_scale": 5.0,
-            "height": 8,
-            "width": 8,
-            "max_sequence_length": 48,
-            "output_type": "np",
-        }
-        return inputs
-
-
-class FluxModularPipelineFastTests(FluxModularTests, ModularPipelineTesterMixin, unittest.TestCase):
-    params = frozenset(["prompt", "height", "width", "guidance_scale"])
-    batch_params = frozenset(["prompt"])
-
-
-class FluxImg2ImgModularPipelineFastTests(FluxModularTests, ModularPipelineTesterMixin, unittest.TestCase):
-    params = frozenset(["prompt", "height", "width", "guidance_scale", "image"])
-    batch_params = frozenset(["prompt", "image"])
-
-    def get_pipeline(self, components_manager=None, torch_dtype=torch.float32):
-        pipeline = super().get_pipeline(components_manager, torch_dtype)
-        # Override `vae_scale_factor` here as currently, `image_processor` is initialized with
-        # fixed constants instead of
-        # https://github.com/huggingface/diffusers/blob/d54622c2679d700b425ad61abce9b80fc36212c0/src/diffusers/pipelines/flux/pipeline_flux_img2img.py#L230C9-L232C10
-        pipeline.image_processor = VaeImageProcessor(vae_scale_factor=2)
-        return pipeline
-
-    def get_dummy_inputs(self, device, seed=0):
-        inputs = super().get_dummy_inputs(device, seed)
-        image = floats_tensor((1, 3, 32, 32), rng=random.Random(seed)).to(device)
-        image = image / 2 + 0.5
-        inputs["image"] = image
-        inputs["strength"] = 0.8
-        inputs["height"] = 8
-        inputs["width"] = 8
-        return inputs
-
-    def test_save_from_pretrained(self):
-        pipes = []
-        base_pipe = self.get_pipeline().to(torch_device)
-        pipes.append(base_pipe)
-
-        with tempfile.TemporaryDirectory() as tmpdirname:
-            base_pipe.save_pretrained(tmpdirname)
-            pipe = ModularPipeline.from_pretrained(tmpdirname).to(torch_device)
-            pipe.load_components(torch_dtype=torch.float32)
-            pipe.to(torch_device)
-            pipe.image_processor = VaeImageProcessor(vae_scale_factor=2)
-
-        pipes.append(pipe)
-
-        image_slices = []
-        for pipe in pipes:
-            inputs = self.get_dummy_inputs(torch_device)
-            image = pipe(**inputs, output="images")
-
-            image_slices.append(image[0, -3:, -3:, -1].flatten())
-
-        assert np.abs(image_slices[0] - image_slices[1]).max() < 1e-3
-
-
-class FluxKontextModularPipelineFastTests(FluxImg2ImgModularPipelineFastTests):
-    pipeline_class = FluxKontextModularPipeline
-    pipeline_blocks_class = FluxKontextAutoBlocks
-    repo = "hf-internal-testing/tiny-flux-kontext-pipe"
-
-    def get_dummy_inputs(self, device, seed=0):
-        inputs = super().get_dummy_inputs(device, seed)
-        image = PIL.Image.new("RGB", (32, 32), 0)
-        _ = inputs.pop("strength")
-        inputs["image"] = image
-        inputs["height"] = 8
-        inputs["width"] = 8
-        inputs["max_area"] = 8 * 8
-        inputs["_auto_resize"] = False
-        return inputs

tests/modular_pipelines/stable_diffusion_xl/test_modular_pipeline_stable_diffusion_xl.py

@@ -21,12 +21,24 @@ import numpy as np
 import torch
 from PIL import Image
 
-from diffusers import ClassifierFreeGuidance, StableDiffusionXLAutoBlocks, StableDiffusionXLModularPipeline
+from diffusers import (
+    ClassifierFreeGuidance,
+    StableDiffusionXLAutoBlocks,
+    StableDiffusionXLModularPipeline,
+)
 from diffusers.loaders import ModularIPAdapterMixin
 
-from ...models.unets.test_models_unet_2d_condition import create_ip_adapter_state_dict
-from ...testing_utils import enable_full_determinism, floats_tensor, torch_device
-from ..test_modular_pipelines_common import ModularPipelineTesterMixin
+from ...models.unets.test_models_unet_2d_condition import (
+    create_ip_adapter_state_dict,
+)
+from ...testing_utils import (
+    enable_full_determinism,
+    floats_tensor,
+    torch_device,
+)
+from ..test_modular_pipelines_common import (
+    ModularPipelineTesterMixin,
+)
 
 
 enable_full_determinism()
@@ -39,7 +51,7 @@ class SDXLModularTests:
 
     pipeline_class = StableDiffusionXLModularPipeline
     pipeline_blocks_class = StableDiffusionXLAutoBlocks
-    repo = "hf-internal-testing/tiny-sdxl-modular"
+    pretrained_model_name_or_path = "hf-internal-testing/tiny-sdxl-modular"
     params = frozenset(
         [
             "prompt",
@@ -54,7 +66,9 @@ class SDXLModularTests:
     batch_params = frozenset(["prompt", "negative_prompt", "image", "mask_image"])
 
     def get_pipeline(self, components_manager=None, torch_dtype=torch.float32):
-        pipeline = self.pipeline_blocks_class().init_pipeline(self.repo, components_manager=components_manager)
+        pipeline = self.pipeline_blocks_class().init_pipeline(
+            self.pretrained_model_name_or_path, components_manager=components_manager
+        )
         pipeline.load_components(torch_dtype=torch_dtype)
         return pipeline
 
@@ -145,7 +159,7 @@ class SDXLModularIPAdapterTests:
 
         blocks = self.pipeline_blocks_class()
         _ = blocks.sub_blocks.pop("ip_adapter")
-        pipe = blocks.init_pipeline(self.repo)
+        pipe = blocks.init_pipeline(self.pretrained_model_name_or_path)
         pipe.load_components(torch_dtype=torch.float32)
         pipe = pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)