update pre-commit config (#18860)

python/sglang/jit_kernel/include/sgl_kernel/type.cuh

@@ -36,27 +36,28 @@ struct dtype_trait {};
   }                                                                 \
   static_assert(true)
 
-SGL_REGISTER_DTYPE_TRAIT(fp32_t, fp32x2_t, SGL_REGISTER_TYPE_END;  //
-                         SGL_REGISTER_FROM_FUNCTION(fp16_t, __half2float);
-                         SGL_REGISTER_FROM_FUNCTION(bf16_t, __bfloat162float);
-                         SGL_REGISTER_UNARY_FUNCTION(abs, fabsf);
-                         SGL_REGISTER_UNARY_FUNCTION(sqrt, sqrtf);
-                         SGL_REGISTER_UNARY_FUNCTION(rsqrt, rsqrtf);
-                         SGL_REGISTER_BINARY_FUNCTION(max, fmaxf);
-                         SGL_REGISTER_BINARY_FUNCTION(min, fminf););
+SGL_REGISTER_DTYPE_TRAIT(
+    fp32_t, fp32x2_t, SGL_REGISTER_TYPE_END;  //
+    SGL_REGISTER_FROM_FUNCTION(fp16_t, __half2float);
+    SGL_REGISTER_FROM_FUNCTION(bf16_t, __bfloat162float);
+    SGL_REGISTER_UNARY_FUNCTION(abs, fabsf);
+    SGL_REGISTER_UNARY_FUNCTION(sqrt, sqrtf);
+    SGL_REGISTER_UNARY_FUNCTION(rsqrt, rsqrtf);
+    SGL_REGISTER_BINARY_FUNCTION(max, fmaxf);
+    SGL_REGISTER_BINARY_FUNCTION(min, fminf););
 SGL_REGISTER_DTYPE_TRAIT(fp16_t, fp16x2_t);
 SGL_REGISTER_DTYPE_TRAIT(bf16_t, bf16x2_t);
 
 /// TODO: Add ROCM implementation
-SGL_REGISTER_DTYPE_TRAIT(fp32x2_t, fp32x4_t, SGL_REGISTER_TYPE_END;
-                         SGL_REGISTER_FROM_FUNCTION(fp16x2_t, __half22float2);
-                         SGL_REGISTER_FROM_FUNCTION(bf16x2_t, __bfloat1622float2););
+SGL_REGISTER_DTYPE_TRAIT(
+    fp32x2_t, fp32x4_t, SGL_REGISTER_TYPE_END; SGL_REGISTER_FROM_FUNCTION(fp16x2_t, __half22float2);
+    SGL_REGISTER_FROM_FUNCTION(bf16x2_t, __bfloat1622float2););
 
-SGL_REGISTER_DTYPE_TRAIT(fp16x2_t, void, SGL_REGISTER_TYPE_END;
-                         SGL_REGISTER_FROM_FUNCTION(fp32x2_t, __float22half2_rn););
+SGL_REGISTER_DTYPE_TRAIT(
+    fp16x2_t, void, SGL_REGISTER_TYPE_END; SGL_REGISTER_FROM_FUNCTION(fp32x2_t, __float22half2_rn););
 
-SGL_REGISTER_DTYPE_TRAIT(bf16x2_t, void, SGL_REGISTER_TYPE_END;
-                         SGL_REGISTER_FROM_FUNCTION(fp32x2_t, __float22bfloat162_rn););
+SGL_REGISTER_DTYPE_TRAIT(
+    bf16x2_t, void, SGL_REGISTER_TYPE_END; SGL_REGISTER_FROM_FUNCTION(fp32x2_t, __float22bfloat162_rn););
 
 #undef SGL_REGISTER_DTYPE_TRAIT
 #undef SGL_REGISTER_FROM_FUNCTION

python/sglang/multimodal_gen/apps/webui/main.py

@@ -243,12 +243,10 @@ def run_sgl_diffusion_webui(server_args: ServerArgs):
         # print banner
         delimiter = "=" * 80
         url = local_url or f"http://localhost:{server_args.webui_port}"
-        print(
-            f"""
+        print(f"""
 {delimiter}
 \033[1mSGLang Diffusion WebUI available at:\033[0m \033[1;4;92m{url}\033[0m
 {delimiter}
-"""
-        )
+""")
 
         demo.block_thread()

python/sglang/multimodal_gen/configs/models/encoders/qwen3.py

@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 """Qwen3 text encoder configuration for SGLang diffusion models."""
+
 from dataclasses import dataclass, field
 
 from sglang.multimodal_gen.configs.models.encoders.base import (

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

python/sglang/multimodal_gen/test/cli/test_generate_common.py

@@ -1,8 +1,9 @@
 # Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
 
 """
-    Common generate cli test, one test for image and video each
+Common generate cli test, one test for image and video each
 """
+
 import dataclasses
 import os
 import shlex

python/sglang/multimodal_gen/test/slack_utils.py

@@ -1,5 +1,5 @@
 """
-    This file upload the media generated in diffusion-nightly-test to a slack channel of SGLang
+This file upload the media generated in diffusion-nightly-test to a slack channel of SGLang
 """
 
 import logging

python/sglang/multimodal_gen/third_party/pynvml.py

@@ -4855,7 +4855,7 @@ def nvmlDeviceGetFieldValues(handle, fieldIds):
 
     for i, fieldId in enumerate(fieldIds):
         try:
-            (values[i].fieldId, values[i].scopeId) = fieldId
+            values[i].fieldId, values[i].scopeId = fieldId
         except TypeError:
             values[i].fieldId = fieldId
 
@@ -4871,7 +4871,7 @@ def nvmlDeviceClearFieldValues(handle, fieldIds):
 
     for i, fieldId in enumerate(fieldIds):
         try:
-            (values[i].fieldId, values[i].scopeId) = fieldId
+            values[i].fieldId, values[i].scopeId = fieldId
         except TypeError:
             values[i].fieldId = fieldId
 

python/sglang/srt/checkpoint_engine/checkpoint_engine_worker.py

@@ -15,6 +15,7 @@
 Checkpoint-engine integration for SGLang.
 This module provides weight update functionality via IPC for checkpoint-engine compatibility.
 """
+
 import logging
 from typing import Callable, Dict, Optional
 

python/sglang/srt/compilation/weak_ref_tensor.py

@@ -13,7 +13,7 @@ else:
 
 
 def weak_ref_tensors(
-    tensors: Union[torch.Tensor, list[torch.Tensor], tuple[torch.Tensor]]
+    tensors: Union[torch.Tensor, list[torch.Tensor], tuple[torch.Tensor]],
 ) -> Union[torch.Tensor, list[Any], tuple[Any], Any]:
     """
     Convenience function to create weak references to tensors,

python/sglang/srt/configs/exaone.py

@@ -14,7 +14,8 @@
 # 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.
-""" EXAONE model configuration """
+"""EXAONE model configuration"""
+
 from typing import Any, Dict
 
 from transformers.configuration_utils import PretrainedConfig

python/sglang/srt/configs/falcon_h1.py

@@ -14,7 +14,6 @@
 # limitations under the License.
 """Falcon-H1 model configuration"""
 
-
 from transformers.configuration_utils import PretrainedConfig
 from transformers.utils import logging
 

python/sglang/srt/disaggregation/mooncake/conn.py

@@ -970,7 +970,7 @@ class MooncakeKVManager(CommonKVManager):
                     self._handle_aux_data(msg)
                     continue
 
-                (bootstrap_room, status, prefill_rank) = msg
+                bootstrap_room, status, prefill_rank = msg
                 status = int(status.decode("ascii"))
                 bootstrap_room = int(bootstrap_room.decode("ascii"))
                 prefill_rank = int(prefill_rank.decode("ascii"))

python/sglang/srt/distributed/parallel_state.py

@@ -21,6 +21,7 @@ If you only need to use the distributed environment without model/pipeline
  parallelism, you can skip the model parallel initialization and destruction
  steps.
 """
+
 import contextlib
 import gc
 import logging
@@ -79,7 +80,7 @@ class P2PWork:
 
 
 def _split_tensor_dict(
-    tensor_dict: Dict[str, Union[torch.Tensor, Any]]
+    tensor_dict: Dict[str, Union[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/srt/entrypoints/openai/serving_responses.py

@@ -1,6 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # Adapted from vLLM's OpenAIServingResponses
 """Handler for /v1/responses requests"""
+
 from __future__ import annotations
 
 import asyncio

python/sglang/srt/function_call/base_format_detector.py

@@ -186,7 +186,7 @@ class BaseFormatDetector(ABC):
                 if start_idx >= len(current_text):
                     return StreamingParseResult()
 
-                (obj, end_idx) = _partial_json_loads(current_text[start_idx:], flags)
+                obj, end_idx = _partial_json_loads(current_text[start_idx:], flags)
 
                 is_current_complete = _is_complete_json(
                     current_text[start_idx : start_idx + end_idx]

python/sglang/srt/hardware_backend/npu/graph_runner/eagle_draft_npu_graph_runner.py

@@ -11,7 +11,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ==============================================================================
-""" Run the model with npu graph and torch.compile """
+"""Run the model with npu graph and torch.compile"""
 
 from __future__ import annotations
 

python/sglang/srt/layers/attention/dual_chunk_flashattention_backend.py

@@ -1,6 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
-"""Attention layer with Dual chunk flash attention and sparse attention.
-"""
+"""Attention layer with Dual chunk flash attention and sparse attention."""
+
 import functools
 import logging
 import math

python/sglang/srt/layers/attention/hybrid_linear_attn_backend.py

@@ -670,9 +670,9 @@ class KimiLinearAttnBackend(MambaAttnBackendBase):
         **kwargs,
     ):
         assert isinstance(mixed_qkv, Tuple)
-        (q_proj_states, k_proj_states, v_proj_states) = mixed_qkv
-        (q_conv_weights, k_conv_weights, v_conv_weights) = layer.conv_weights
-        (q_conv_bias, k_conv_bias, v_conv_bias) = layer.bias
+        q_proj_states, k_proj_states, v_proj_states = mixed_qkv
+        q_conv_weights, k_conv_weights, v_conv_weights = layer.conv_weights
+        q_conv_bias, k_conv_bias, v_conv_bias = layer.bias
 
         layer_cache = self.req_to_token_pool.mamba2_layer_cache(layer.layer_id)
         q_conv_state, k_conv_state, v_conv_state = layer_cache.conv
@@ -744,9 +744,9 @@ class KimiLinearAttnBackend(MambaAttnBackendBase):
         )
 
         assert isinstance(mixed_qkv, Tuple)
-        (q_proj_states, k_proj_states, v_proj_states) = mixed_qkv
-        (q_conv_weights, k_conv_weights, v_conv_weights) = layer.conv_weights
-        (q_conv_bias, k_conv_bias, v_conv_bias) = layer.bias
+        q_proj_states, k_proj_states, v_proj_states = mixed_qkv
+        q_conv_weights, k_conv_weights, v_conv_weights = layer.conv_weights
+        q_conv_bias, k_conv_bias, v_conv_bias = layer.bias
 
         query_start_loc = self.forward_metadata.query_start_loc
         cache_indices = self.forward_metadata.mamba_cache_indices

python/sglang/srt/layers/attention/linear/seg_la.py

@@ -2,6 +2,7 @@
 """
 Copyright (c) Ant Financial Service Group and its affiliates.
 """
+
 # Copied from https://code.alipay.com/pia/PainlessInferenceAcceleration/blob/v0.0.6/flood/flood/ops/seg_la.py
 
 from dataclasses import dataclass

python/sglang/srt/layers/attention/wave_ops/prefill_attention.py

@@ -38,7 +38,7 @@ def prefill_attention_wave(
     output_shape = (shape.total_seq_len, shape.num_query_heads, shape.head_size_kv)
     # Run the wave kernel.
     mfma_variant = (MMAType.F32_16x16x16_F16, MMAType.F32_16x16x16_F16)
-    (prefill, hyperparams) = get_prefill_attention_kernel(
+    prefill, hyperparams = get_prefill_attention_kernel(
         shape,
         mfma_variant,
         q.shape,

python/sglang/srt/layers/logits_processor.py

@@ -376,7 +376,7 @@ class LogitsProcessor(nn.Module):
 
             logprobs_result = self.process_input_logprobs(input_logits, logits_metadata)
         else:
-            (logprobs_result, sampled_logits) = self.process_input_logprobs_by_chunk(
+            logprobs_result, sampled_logits = self.process_input_logprobs_by_chunk(
                 pruned_states,
                 sample_indices,
                 input_logprob_indices,

python/sglang/srt/layers/moe/cutlass_w4a8_moe.py

@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 """Cutlass W4A8 MoE kernel."""
+
 from typing import Optional
 
 import torch

python/sglang/srt/layers/quantization/modelslim/modelslim_moe.py

@@ -63,11 +63,9 @@ class ModelSlimMoEMethod(FusedMoEMethodBase):
             logger.info_once("Using ModelSlimW8A8Int8MoE")
             return ModelSlimW8A8Int8MoE(quant_config)
         else:
-            logger.warning(
-                f"Unsupported FusedMoe modelslim scheme: \
+            logger.warning(f"Unsupported FusedMoe modelslim scheme: \
                     {quant_config.quant_description.get(prefix_in_quant_config.strip())} \
-                    in layer: {prefix}"
-            )
+                    in layer: {prefix}")
             return None
 
 

python/sglang/srt/layers/radix_attention.py

@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """Radix attention."""
+
 from __future__ import annotations
 
 from enum import Enum

python/sglang/srt/layers/radix_linear_attention.py

@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """Radix linear attention."""
+
 from __future__ import annotations
 
 from typing import TYPE_CHECKING, Optional, Tuple, Union

python/sglang/srt/layers/rotary_embedding.py

@@ -1,5 +1,6 @@
 # Adapted from https://raw.githubusercontent.com/vllm-project/vllm/refs/tags/v0.6.6.post1/vllm/model_executor/layers/rotary_embedding.py
 """Rotary Positional Embeddings."""
+
 from __future__ import annotations
 
 import itertools
@@ -2896,7 +2897,7 @@ class DualChunkRotaryEmbedding(MultiPlatformOp):
         self.local_size = local_size
         self.dtype = dtype
         self.device = torch.device(f"cuda:{torch.cuda.current_device()}")
-        (q_cache, qc_cache, k_cache, qc_no_clamp_cache, q_inter_cache) = (
+        q_cache, qc_cache, k_cache, qc_no_clamp_cache, q_inter_cache = (
             self._compute_cos_sin_cache()
         )
 

python/sglang/srt/managers/tp_worker.py

@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """A tensor parallel worker."""
+
 from __future__ import annotations
 
 import logging

python/sglang/srt/mem_cache/mamba_radix_cache.py

@@ -639,7 +639,7 @@ class MambaRadixCache(BasePrefixCache):
         match_result = self.match_prefix(
             MatchPrefixParams(key=RadixKey(page_aligned_token_ids, req.extra_key))
         )
-        (new_indices, new_last_node) = (
+        new_indices, new_last_node = (
             match_result.device_indices,
             match_result.last_device_node,
         )

python/sglang/srt/mem_cache/memory_pool_host.py

@@ -8,7 +8,9 @@ from typing import Optional
 import psutil
 import torch
 
-from sglang.jit_kernel.hicache import can_use_hicache_jit_kernel
+from sglang.jit_kernel.hicache import (
+    can_use_hicache_jit_kernel,
+)
 from sglang.jit_kernel.hicache import (
     transfer_hicache_all_layer as jit_transfer_hicache_all_layer,
 )

python/sglang/srt/mem_cache/radix_cache.py

@@ -523,7 +523,7 @@ class RadixCache(BasePrefixCache):
 
         # The prefix indices could be updated, reuse it
         match_result = self.match_prefix(MatchPrefixParams(key=radix_key))
-        (new_indices, new_last_node) = (
+        new_indices, new_last_node = (
             match_result.device_indices,
             match_result.last_device_node,
         )

python/sglang/srt/mem_cache/swa_radix_cache.py

@@ -556,7 +556,7 @@ class SWARadixCache(BasePrefixCache):
         match_result = self.match_prefix(
             MatchPrefixParams(key=RadixKey(page_aligned_token_ids, req.extra_key))
         )
-        (new_indices, new_last_node) = (
+        new_indices, new_last_node = (
             match_result.device_indices,
             match_result.last_device_node,
         )

python/sglang/srt/metrics/collector.py

@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """Utilities for Prometheus Metrics Collection."""
+
 import dataclasses
 import logging
 import os

python/sglang/srt/metrics/utils.py

@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """Utilities for Prometheus Metrics."""
+
 import math
 from typing import List
 

python/sglang/srt/model_loader/utils.py

@@ -1,6 +1,7 @@
 # Adapted from https://github.com/vllm-project/vllm/blob/v0.6.4.post1/vllm/model_executor/model_loader/utils.py
 
 """Utilities for selecting and loading models."""
+
 import concurrent.futures
 import contextlib
 import logging

python/sglang/srt/model_loader/weight_utils.py

@@ -1,6 +1,7 @@
 # Adapted from https://github.com/vllm-project/vllm/blob/v0.6.4.post1/vllm/model_executor/model_loader/weight_utils.py
 
 """Utilities for downloading and initializing model weights."""
+
 import collections
 import concurrent.futures
 import fnmatch
@@ -140,12 +141,10 @@ def convert_bin_to_safetensor_file(
     sf_size = os.stat(sf_filename).st_size
     pt_size = os.stat(pt_filename).st_size
     if (sf_size - pt_size) / pt_size > 0.01:
-        raise RuntimeError(
-            f"""The file size different is more than 1%:
+        raise RuntimeError(f"""The file size different is more than 1%:
          - {sf_filename}: {sf_size}
          - {pt_filename}: {pt_size}
-         """
-        )
+         """)
 
     # check if the tensors are the same
     reloaded = safetensors.torch.load_file(sf_filename)

python/sglang/srt/models/baichuan.py

@@ -18,6 +18,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only BaiChuan model compatible with HuggingFace weights."""
+
 import math
 from typing import Iterable, Optional, Tuple
 

python/sglang/srt/models/bailing_moe.py

@@ -18,6 +18,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """SGLang BailingMoE model."""
+
 import logging
 from typing import Iterable, List, Optional, Tuple, Union
 

python/sglang/srt/models/bailing_moe_nextn.py

@@ -18,6 +18,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """SGLang BailingMoENextN model."""
+
 import logging
 from typing import Iterable, Optional, Tuple
 

python/sglang/srt/models/deepseek_nextn.py

@@ -13,6 +13,7 @@
 # ==============================================================================
 
 """Inference-only DeepSeek NextN Speculative Decoding."""
+
 import logging
 from typing import Iterable, Optional, Tuple
 

python/sglang/srt/models/deepseek_ocr.py

@@ -16,6 +16,7 @@
 # Adapted from
 # https://github.com/vllm-project/vllm/blob/c7f2cf2b7f67bce5842fedfdba508440fe257375/vllm/model_executor/models/llama.py#L1
 """Inference-only Apertus model compatible with HuggingFace weights."""
+
 import copy
 import logging
 import math

python/sglang/srt/models/deepseek_v2.py

@@ -15,6 +15,7 @@
 # Adapted from:
 # https://github.com/vllm-project/vllm/blob/fb6af8bc086328ca6659e72d11ffd4309ce4de22/vllm/model_executor/models/deepseek_v2.py
 """Inference-only DeepseekV2 model."""
+
 from __future__ import annotations
 
 import logging

python/sglang/srt/models/ernie4.py

@@ -12,7 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 
-""" Inference-only Ernie4.5 model compatible with baidu/ERNIE-4.5-*-PT weights. """
+"""Inference-only Ernie4.5 model compatible with baidu/ERNIE-4.5-*-PT weights."""
 
 from typing import Iterable, List, Optional, Tuple, Union
 

python/sglang/srt/models/ernie45_moe_vl.py

@@ -12,7 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 
-""" Inference-only Ernie4.5 VL model compatible with baidu/ERNIE-4.5-VL-*-PT weights. """
+"""Inference-only Ernie4.5 VL model compatible with baidu/ERNIE-4.5-VL-*-PT weights."""
 
 import logging
 from itertools import islice

python/sglang/srt/models/ernie45_vl.py

@@ -11,6 +11,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only Ernie45-VL model compatible with HuggingFace weights."""
+
 import logging
 from functools import lru_cache, partial
 from typing import Iterable, List, Optional, Tuple, Type

python/sglang/srt/models/ernie4_eagle.py

@@ -12,7 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 
-""" Ernie4.5 MTP model compatible with baidu/ERNIE-4.5-*-PT weights. """
+"""Ernie4.5 MTP model compatible with baidu/ERNIE-4.5-*-PT weights."""
 
 from typing import Iterable, Optional, Tuple
 

python/sglang/srt/models/gpt2.py

@@ -17,6 +17,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only GPT-2 model compatible with HuggingFace weights."""
+
 from typing import Iterable, Optional, Tuple, Type
 
 import torch

python/sglang/srt/models/hunyuan.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only HunYuan model compatible with HuggingFace weights."""
+
 import re
 from typing import Any, Dict, Iterable, Optional, Tuple
 

python/sglang/srt/models/llada2.py

@@ -18,6 +18,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """SGLang LLaDA2MoeModelLM model."""
+
 import logging
 from typing import Iterable, Optional, Tuple, Union
 

python/sglang/srt/models/mllama.py

@@ -1,6 +1,7 @@
 # Adapted from:
 # https://github.com/vllm-project/vllm/blob/7193774b1ff8603ad5bf4598e5efba0d9a39b436/vllm/model_executor/models/mllama.py
 """PyTorch Mllama model."""
+
 import math
 from typing import Iterable, List, Optional, Tuple, Union
 

python/sglang/srt/models/nemotron_nas.py

@@ -14,6 +14,7 @@
 # Adapted from https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/nemotron_nas.py
 
 """Inference-only deci model compatible with HuggingFace weights."""
+
 from typing import Iterable, Optional, Tuple, Type, Union
 
 import torch

python/sglang/srt/models/olmo.py

@@ -15,6 +15,7 @@
 # Adapted from
 # https://github.com/vllm-project/vllm/blob/c7f2cf2b7f67bce5842fedfdba508440fe257375/vllm/model_executor/models/olmo.py#L1
 """Inference-only OLMo model compatible with HuggingFace weights."""
+
 from typing import Iterable, Optional, Tuple
 
 import torch

python/sglang/srt/models/olmo2.py

@@ -15,6 +15,7 @@
 # Adapted from
 # https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/olmo2.py
 """Inference-only OLMo2 model compatible with HuggingFace weights."""
+
 from functools import partial
 from typing import Iterable, Optional, Tuple
 

python/sglang/srt/models/opt.py

@@ -13,6 +13,7 @@
 # ==============================================================================
 
 """Inference-only OPT model compatible with HuggingFace weights."""
+
 import logging
 from collections.abc import Iterable
 from typing import Optional, Union

python/sglang/srt/models/orion.py

@@ -7,6 +7,7 @@
 # LICENSE: https://huggingface.co/OrionStarAI/Orion-14B-Base/blob/main/LICENSE
 # Adapted from https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/orion.py
 """Inference-only Orion-14B model compatible with HuggingFace weights."""
+
 from collections.abc import Iterable
 from typing import Any, Optional, Tuple
 

python/sglang/srt/models/pixtral.py

@@ -23,11 +23,15 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 from transformers import PixtralVisionConfig, PretrainedConfig
-from transformers.models.pixtral.modeling_pixtral import PixtralRotaryEmbedding
+from transformers.models.pixtral.modeling_pixtral import (
+    PixtralRotaryEmbedding,
+)
 from transformers.models.pixtral.modeling_pixtral import (
     generate_block_attention_mask as _get_pixtral_attention_mask,
 )
-from transformers.models.pixtral.modeling_pixtral import position_ids_in_meshgrid
+from transformers.models.pixtral.modeling_pixtral import (
+    position_ids_in_meshgrid,
+)
 
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.attention.vision import VisionAttention

python/sglang/srt/models/qwen2.py

@@ -15,6 +15,7 @@
 # Adapted from llama2.py
 # Modify details for the adaptation of Qwen2 model.
 """Inference-only Qwen2 model compatible with HuggingFace weights."""
+
 import logging
 from typing import Any, Dict, Iterable, List, Optional, Tuple, Union
 

python/sglang/srt/models/qwen2_5_vl.py

@@ -22,6 +22,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only Qwen2-VL model compatible with HuggingFace weights."""
+
 import logging
 import re
 from functools import partial

python/sglang/srt/models/qwen2_audio.py

@@ -22,6 +22,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only Qwen2-Audio model compatible with HuggingFace weights."""
+
 import logging
 from typing import Any, Iterable, List, Optional, Tuple
 

python/sglang/srt/models/qwen2_vl.py

@@ -22,6 +22,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only Qwen2-VL model compatible with HuggingFace weights."""
+
 import logging
 from functools import lru_cache, partial
 from typing import Iterable, List, Optional, Tuple, Type, TypedDict

python/sglang/srt/models/qwen3_5.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 # ==============================================================================
 """Inference-only Qwen3.5 model and Qwen3.5 MoE model compatible with HuggingFace weights."""
+
 import logging
 from functools import lru_cache
 from typing import Iterable, Optional, Set, Tuple, Union

python/sglang/srt/models/qwen3_5_mtp.py

@@ -13,6 +13,7 @@
 # ==============================================================================
 
 """Inference-only Qwen3_5 MTP model."""
+
 import logging
 from typing import Iterable, Optional, Tuple
 

python/sglang/srt/models/qwen3_next.py

@@ -366,8 +366,8 @@ class Qwen3GatedDeltaNet(nn.Module):
 
         # [b, sq, ng, (hn + hn + np/ng * hn + np/ng + np/ng)]
         # --> [b, sq, ng, hn], [b, sq, ng, hn], [b, sq, ng, np/ng * hn], [b, sq, ng, np/ng * hn], [b, sq, ng, np/ng], [b, sq, ng, np/ng]
-        (query, key, value, z) = torch.split(mixed_qkvz, split_arg_list_qkvz, dim=2)
-        (b, a) = torch.split(mixed_ba, split_arg_list_ba, dim=2)
+        query, key, value, z = torch.split(mixed_qkvz, split_arg_list_qkvz, dim=2)
+        b, a = torch.split(mixed_ba, split_arg_list_ba, dim=2)
 
         # [b, sq, ng, np/ng * hn] -> [b, sq, np, hn]
         value = value.reshape(value.size(0), -1, self.head_v_dim)

python/sglang/srt/models/qwen3_next_mtp.py

@@ -13,6 +13,7 @@
 # ==============================================================================
 
 """Inference-only Qwen3Next MTP Speculative Decoding."""
+
 import logging
 from typing import Iterable, Optional, Tuple
 

python/sglang/srt/models/qwen3_omni_moe.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 # ==============================================================================
 """Inference-only Qwen3-VL model compatible with HuggingFace weights."""
+
 import math
 from typing import Iterable, List, Optional, Tuple
 

python/sglang/srt/models/qwen3_vl.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 # ==============================================================================
 """Inference-only Qwen3-VL model compatible with HuggingFace weights."""
+
 import logging
 import math
 import re

python/sglang/srt/models/qwen3_vl_moe.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 # ==============================================================================
 """Inference-only Qwen3-VL model compatible with HuggingFace weights."""
+
 import logging
 import re
 from functools import lru_cache

python/sglang/srt/models/starcoder2.py

@@ -20,7 +20,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # Adapted from https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/starcoder2.py
-""" PyTorch Starcoder2 model."""
+"""PyTorch Starcoder2 model."""
+
 from collections.abc import Iterable
 from typing import Optional, Tuple
 

python/sglang/srt/models/transformers.py

@@ -15,6 +15,7 @@
 # Adapted from
 # https://github.com/vllm-project/vllm/blob/a1a2aaadb9122f05667140e39cf67e5736c8b6d6/vllm/model_executor/models/transformers.py
 """Wrapper around `transformers` models"""
+
 import logging
 import re
 from typing import Iterable, Literal, Optional, Tuple, Union

python/sglang/srt/multimodal/internvl_vit_cuda_graph_runner.py

@@ -13,6 +13,7 @@
 # ==============================================================================
 
 """ViT CUDA Graph Runner class."""
+
 from __future__ import annotations
 
 from typing import Dict, Hashable, Tuple

python/sglang/srt/multimodal/mm_utils.py

@@ -27,6 +27,7 @@ LLaVA-NeXT : https://llava-vl.github.io/blog/2024-01-30-llava-next/
 LLaVA-Onevision : https://arxiv.org/pdf/2408.03326
 
 """
+
 import ast
 import itertools
 import math
@@ -519,7 +520,7 @@ def run_dp_sharded_mrope_vision_model(
     # image_to_tp_rank = [0, 2, 1, 3]
     # gpu_sample_counts = [1, 3]
     # grouped_pixel_values_len = [1000, 350]
-    (image_to_tp_rank, gpu_sample_counts, grouped_pixel_values_len) = (
+    image_to_tp_rank, gpu_sample_counts, grouped_pixel_values_len = (
         get_dp_encoder_lb_assignment(patches_per_image, tp_size)
     )
 

python/sglang/srt/multimodal/processors/ernie45_vl.py

@@ -15,7 +15,9 @@ from sglang.srt.models.ernie45_vl import Ernie4_5_VLMoeForConditionalGeneration
 from sglang.srt.multimodal.processors.base_processor import (
     BaseMultimodalProcessor as SGLangBaseProcessor,
 )
-from sglang.srt.multimodal.processors.base_processor import MultimodalSpecialTokens
+from sglang.srt.multimodal.processors.base_processor import (
+    MultimodalSpecialTokens,
+)
 from sglang.srt.utils import get_bool_env_var, is_npu, logger
 
 _is_npu = is_npu()

python/sglang/srt/multimodal/processors/glm4v.py

@@ -7,7 +7,9 @@ from sglang.srt.models.glm_ocr import GlmOcrForConditionalGeneration
 from sglang.srt.multimodal.processors.base_processor import (
     BaseMultimodalProcessor as SGLangBaseProcessor,
 )
-from sglang.srt.multimodal.processors.base_processor import MultimodalSpecialTokens
+from sglang.srt.multimodal.processors.base_processor import (
+    MultimodalSpecialTokens,
+)
 
 
 class Glm4vImageProcessor(SGLangBaseProcessor):

python/sglang/srt/multimodal/processors/kimi_k25.py

@@ -8,7 +8,9 @@ from sglang.srt.models.kimi_k25 import KimiK25ForConditionalGeneration
 from sglang.srt.multimodal.processors.base_processor import (
     BaseMultimodalProcessor as SGLangBaseProcessor,
 )
-from sglang.srt.multimodal.processors.base_processor import MultimodalSpecialTokens
+from sglang.srt.multimodal.processors.base_processor import (
+    MultimodalSpecialTokens,
+)
 
 
 # Compatible with KimiVLForConditionalGeneration

python/sglang/srt/multimodal/processors/kimi_vl.py

@@ -5,7 +5,9 @@ from sglang.srt.models.kimi_vl import KimiVLForConditionalGeneration
 from sglang.srt.multimodal.processors.base_processor import (
     BaseMultimodalProcessor as SGLangBaseProcessor,
 )
-from sglang.srt.multimodal.processors.base_processor import MultimodalSpecialTokens
+from sglang.srt.multimodal.processors.base_processor import (
+    MultimodalSpecialTokens,
+)
 
 
 # Compatible with KimiVLForConditionalGeneration

python/sglang/srt/multimodal/processors/qwen_vl.py

@@ -26,7 +26,9 @@ from sglang.srt.models.qwen3_vl_moe import Qwen3VLMoeForConditionalGeneration
 from sglang.srt.multimodal.processors.base_processor import (
     BaseMultimodalProcessor as SGLangBaseProcessor,
 )
-from sglang.srt.multimodal.processors.base_processor import MultimodalSpecialTokens
+from sglang.srt.multimodal.processors.base_processor import (
+    MultimodalSpecialTokens,
+)
 from sglang.utils import logger
 
 IMAGE_FACTOR = 28

python/sglang/srt/multimodal/processors/step3_vl.py

@@ -15,7 +15,9 @@ from sglang.srt.models.step3_vl_10b import StepVLForConditionalGeneration
 from sglang.srt.multimodal.processors.base_processor import (
     BaseMultimodalProcessor as SGLangBaseProcessor,
 )
-from sglang.srt.multimodal.processors.base_processor import MultimodalSpecialTokens
+from sglang.srt.multimodal.processors.base_processor import (
+    MultimodalSpecialTokens,
+)
 
 ImageWithPatches = tuple[Image.Image, list[Image.Image], list[int] | None]
 

python/sglang/srt/multimodal/vit_cuda_graph_runner.py

@@ -13,6 +13,7 @@
 # ==============================================================================
 
 """ViT CUDA Graph Runner class."""
+
 from __future__ import annotations
 
 import inspect

python/sglang/srt/parser/code_completion_parser.py

@@ -13,7 +13,6 @@
 # ==============================================================================
 """Completion templates."""
 
-
 import dataclasses
 import logging
 from enum import auto

python/sglang/srt/utils/common.py

@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """Common utilities."""
+
 from __future__ import annotations
 
 import argparse

python/sglang/srt/utils/weight_checker.py

@@ -117,7 +117,7 @@ def _random_like(t: torch.Tensor):
 
 
 def _postprocess_tensors(
-    raw: Dict[str, torch.Tensor]
+    raw: Dict[str, torch.Tensor],
 ) -> Iterable[Tuple[str, bool, torch.Tensor]]:
     from sglang.srt.debug_utils.dumper import get_tensor_info
 

python/sglang/utils.py

@@ -531,15 +531,13 @@ def wait_for_server(
         headers={"Authorization": "Bearer None"},
     )
     time.sleep(5)
-    print_highlight(
-        """\n
+    print_highlight("""\n
         NOTE: Typically, the server runs in a separate terminal.
         In this notebook, we run the server and notebook code together, so their outputs are combined.
         To improve clarity, the server logs are displayed in the original black color, while the notebook outputs are highlighted in blue.
         To reduce the log length, we set the log level to warning for the server, the default log level is info.
         We are running those notebooks in a CI environment, so the throughput is not representative of the actual performance.
-        """
-    )
+        """)
 
 
 class TypeBasedDispatcher: