update pre-commit config (#18860)

python/sglang/multimodal_gen/runtime/cache/__init__.py

@@ -9,6 +9,7 @@ diffusion transformer (DiT) inference:
 - cache-dit integration: Block-level caching with DBCache and TaylorSeer
 
 """
+
 from sglang.multimodal_gen.runtime.cache.cache_dit_integration import (
     CacheDitConfig,
     enable_cache_on_dual_transformer,

python/sglang/multimodal_gen/runtime/distributed/parallel_state.py

@@ -29,6 +29,7 @@ The typical workflow is:
 If you only need to use the distributed environment without model parallelism,
  you can skip the model parallel initialization and destruction steps.
 """
+
 import contextlib
 import datetime
 import os
@@ -71,7 +72,7 @@ TensorMetadata = namedtuple("TensorMetadata", ["device", "dtype", "size"])
 
 
 def _split_tensor_dict(
-    tensor_dict: dict[str, torch.Tensor | Any]
+    tensor_dict: dict[str, torch.Tensor | Any],
 ) -> tuple[list[tuple[str, Any]], list[torch.Tensor]]:
     """Split the tensor dictionary into two parts:
     1. A list of (key, value) pairs. If the value is a tensor, it is replaced

python/sglang/multimodal_gen/runtime/layers/activation.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # Adapted from vllm: https://github.com/vllm-project/vllm/blob/v0.7.3/vllm/model_executor/layers/activation.py
 """Custom activation functions."""
+
 import math
 from typing import Any
 

python/sglang/multimodal_gen/runtime/layers/layernorm.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # Adapted from vllm: https://github.com/vllm-project/vllm/blob/v0.7.3/vllm/model_executor/layers/layernorm.py
 """Custom normalization layers."""
+
 from typing import Optional, Tuple, Union
 
 import torch

python/sglang/multimodal_gen/runtime/layers/rotary_embedding.py

@@ -25,6 +25,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Rotary Positional Embeddings."""
+
 import functools
 from collections import OrderedDict
 from typing import Any, Optional, Tuple

python/sglang/multimodal_gen/runtime/layers/utils.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # Adapted from vllm: https://github.com/vllm-project/vllm/blob/v0.7.3/vllm/model_executor/layers/utils.py
 """Utility methods for model layers."""
+
 import inspect
 from typing import Any, Callable, List, Optional
 

python/sglang/multimodal_gen/runtime/layers/visual_embedding.py

@@ -12,7 +12,10 @@ from diffusers.models.embeddings import (
 from diffusers.models.embeddings import (
     CombinedTimestepTextProjEmbeddings as _CombinedTimestepTextProjEmbeddings,
 )
-from diffusers.models.embeddings import PixArtAlphaTextProjection, TimestepEmbedding
+from diffusers.models.embeddings import (
+    PixArtAlphaTextProjection,
+    TimestepEmbedding,
+)
 from diffusers.models.embeddings import Timesteps as _Timesteps
 from diffusers.models.embeddings import (
     get_timestep_embedding as timestep_embedding_diffusers,

python/sglang/multimodal_gen/runtime/loader/utils.py

@@ -2,6 +2,7 @@
 
 # SPDX-License-Identifier: Apache-2.0
 """Utilities for selecting and loading models."""
+
 import contextlib
 import glob
 import os
@@ -30,7 +31,7 @@ def set_default_torch_dtype(dtype: torch.dtype):
 
 
 def get_param_names_mapping(
-    mapping_dict: dict[str, str]
+    mapping_dict: dict[str, str],
 ) -> Callable[[str], tuple[str, Any, Any]]:
     """
     Creates a mapping function that transforms parameter names using regex patterns.

python/sglang/multimodal_gen/runtime/loader/weight_utils.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # Adapted from vllm: https://github.com/vllm-project/vllm/blob/v0.7.3/vllm/model_executor/model_loader/weight_utils.py
 """Utilities for downloading and initializing model weights."""
+
 import hashlib
 import json
 import os

python/sglang/multimodal_gen/runtime/models/encoders/clip.py

@@ -5,6 +5,7 @@
 # Adapted from transformers: https://github.com/huggingface/transformers/blob/v4.39.0/src/transformers/models/clip/modeling_clip.py
 """Minimal implementation of CLIPVisionModel intended to be only used
 within a vision language model."""
+
 from collections.abc import Iterable
 from typing import Optional
 

python/sglang/multimodal_gen/runtime/models/encoders/llama.py

@@ -25,6 +25,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only LLaMA model compatible with HuggingFace weights."""
+
 from collections.abc import Iterable
 from typing import Any
 

python/sglang/multimodal_gen/runtime/models/utils.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # Adapted from: https://github.com/vllm-project/vllm/blob/v0.7.3/vllm/model_executor/utils.py
 """Utils for model executor."""
+
 from typing import Any
 
 import torch

python/sglang/multimodal_gen/runtime/pipelines/hunyuan_pipeline.py

@@ -8,7 +8,6 @@ This module contains an implementation of the Hunyuan video diffusion pipeline
 using the modular pipeline architecture.
 """
 
-
 from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )

python/sglang/multimodal_gen/runtime/pipelines_core/executors/sync_executor.py

@@ -4,6 +4,7 @@
 """
 Synchronous pipeline executor implementation.
 """
+
 from typing import List
 
 from sglang.multimodal_gen.runtime.pipelines_core.executors.pipeline_executor import (

python/sglang/multimodal_gen/runtime/pipelines_core/stages/comfyui_latent_preparation.py

@@ -5,6 +5,7 @@ This stage extends LatentPreparationStage to handle device mismatch issues
 that occur when tensors are pickled and unpickled via broadcast_pyobj in
 multi-GPU scenarios.
 """
+
 import dataclasses
 
 import torch

python/sglang/multimodal_gen/runtime/pipelines_core/stages/input_validation.py

@@ -4,6 +4,7 @@
 """
 Input validation stage for diffusion pipelines.
 """
+
 import numpy as np
 import torch
 import torchvision.transforms.functional as TF

python/sglang/multimodal_gen/runtime/pipelines_core/stages/latent_preparation.py

@@ -4,6 +4,7 @@
 """
 Latent preparation stage for diffusion pipelines.
 """
+
 from diffusers.utils.torch_utils import randn_tensor
 
 from sglang.multimodal_gen.runtime.distributed import get_local_torch_device

python/sglang/multimodal_gen/runtime/platforms/rocm.py

@@ -6,6 +6,7 @@
 This file is a platform abstraction for ROCm GPUs,
 adjusted to match the structure and interface of `cuda.py`.
 """
+
 from functools import lru_cache
 from typing import Any
 

python/sglang/multimodal_gen/runtime/server_args.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # Inspired by SGLang: https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/server_args.py
 """The arguments of sglang-diffusion Inference."""
+
 import argparse
 import dataclasses
 import inspect

python/sglang/multimodal_gen/runtime/utils/logging_utils.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # adapted from vllm: https://github.com/vllm-project/vllm/blob/v0.7.3/vllm/logger.py
 """Logging configuration for sglang.multimodal_gen."""
+
 import argparse
 import contextlib
 import datetime