update pre-commit config (#18860)

2026-02-16 00:18:31 +08:00
parent f7603203b0
commit 07a24f1a38
135 changed files with 239 additions and 198 deletions
@@ -26,9 +26,9 @@ jobs:
        run: SKIP=no-commit-to-branch pre-commit run --all-files --show-diff-on-failure
      - name: Run sgl-kernel clang-format checks
-        uses: DoozyX/clang-format-lint-action@v0.18.1
+        uses: DoozyX/clang-format-lint-action@v0.20
        with:
          source: sgl-kernel
          extensions: h,c,cpp,hpp,cu,cuh,cc
-          clangFormatVersion: 18
+          clangFormatVersion: 20
          style: file
@@ -3,7 +3,7 @@ exclude: ^(python/sglang/multimodal_gen/csrc|python/sglang/jit_kernel/flash_atte
 repos:
  - repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v5.0.0
+    rev: v6.0.0
    hooks:
      - id: check-symlinks
      - id: destroyed-symlinks
@@ -21,12 +21,12 @@ repos:
      - id: debug-statements
      - id: no-commit-to-branch
  - repo: https://github.com/PyCQA/isort
-    rev: 5.13.2
+    rev: 7.0.0
    hooks:
      - id: isort
        exclude: '^python/sglang/srt/grpc/.*_pb2\.py$|^python/sglang/srt/grpc/.*_pb2_grpc\.py$|^python/sglang/srt/grpc/.*_pb2\.pyi$|^python/sglang/srt/grpc/.*_pb2_grpc\.pyi$'
  - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.11.7
+    rev: v0.15.1
    hooks:
      - id: ruff
        args:
@@ -43,7 +43,7 @@ repos:
          python/sglang/srt/grpc/.*_pb2_grpc\.pyi$|
          )$
  - repo: https://github.com/psf/black
-    rev: 24.10.0
+    rev: 26.1.0
    hooks:
      - id: black-jupyter
        exclude: '^python/sglang/srt/grpc/.*_pb2\.py$|^python/sglang/srt/grpc/.*_pb2_grpc\.py$|^python/sglang/srt/grpc/.*_pb2\.pyi$|^python/sglang/srt/grpc/.*_pb2_grpc\.pyi$'
@@ -53,13 +53,13 @@ repos:
      - id: codespell
        args: ['--config', '.codespellrc']
  - repo: https://github.com/pre-commit/mirrors-clang-format
-    rev: v18.1.8
+    rev: v20.1.7
    hooks:
    - id: clang-format
      types_or: [c++, cuda]
      args: [--style=file, --verbose]
  - repo: https://github.com/kynan/nbstripout
-    rev: 0.8.1
+    rev: 0.9.0
    hooks:
      - id: nbstripout
        args:
@@ -187,10 +187,8 @@ def run_grid(bs, model, method, tp_size, dtype: str):
    configs = union_of_list_of_dicts(prune_configs_1, prune_configs_2)
-    print(
+    print(f"{bs=} || {len(full_configs)=} | {len(prune_configs_1)=} | \
-        f"{bs=} || {len(full_configs)=} | {len(prune_configs_1)=} | \
+            {len(prune_configs_2)=} | {len(configs)=}")
            {len(prune_configs_2)=} | {len(configs)=}"
    )
    best_config = None
    best_time_us = 1e20
@@ -7,7 +7,9 @@ import torch
 from sglang.srt.layers.attention.fla.layernorm_gated import (
    _layer_norm_fwd as layer_norm_fwd,
 )
-from sglang.srt.layers.attention.fla.layernorm_gated import rms_norm_ref
+from sglang.srt.layers.attention.fla.layernorm_gated import (
    rms_norm_ref,
 )
 def benchmark_layer_norm_fwd(
@@ -13,8 +13,7 @@ number = 5
 def expand_tip(topic, tip, generate):
-    s = (
+    s = """Please expand a tip for a topic into a detailed paragraph.
        """Please expand a tip for a topic into a detailed paragraph.
 Topic: staying healthy
 Tip: Regular Exercise
@@ -28,12 +27,7 @@ Topic: writing a blog post
 Tip: structure your content effectively
 Paragraph: A well-structured post is easier to read and more enjoyable. Start with an engaging introduction that hooks the reader and clearly states the purpose of your post. Use headings and subheadings to break up the text and guide readers through your content. Bullet points and numbered lists can make information more digestible. Ensure each paragraph flows logically into the next, and conclude with a summary or call-to-action that encourages reader engagement.
-Topic: """
+Topic: """ + topic + "\nTip: " + tip + "\nParagraph:"
        + topic
        + "\nTip: "
        + tip
        + "\nParagraph:"
    )
    return generate(s, max_tokens=128, stop=["\n\n"])
@@ -14,8 +14,7 @@ number = 5
@sgl.function
 def expand_tip(s, topic, tip):
-    s += (
+    s += """Please expand a tip for a topic into a detailed paragraph.
        """Please expand a tip for a topic into a detailed paragraph.
 Topic: staying healthy
 Tip: Regular Exercise
@@ -29,12 +28,7 @@ Topic: writing a blog post
 Tip: structure your content effectively
 Paragraph: A well-structured post is easier to read and more enjoyable. Start with an engaging introduction that hooks the reader and clearly states the purpose of your post. Use headings and subheadings to break up the text and guide readers through your content. Bullet points and numbered lists can make information more digestible. Ensure each paragraph flows logically into the next, and conclude with a summary or call-to-action that encourages reader engagement.
-Topic: """
+Topic: """ + topic + "\nTip: " + tip + "\nParagraph:"
        + topic
        + "\nTip: "
        + tip
        + "\nParagraph:"
    )
    s += sgl.gen("paragraph", max_tokens=128, stop=["\n\n"], temperature=0)
@@ -2,8 +2,7 @@ number = 5
 async def expand_tip_async(topic, tip, generate):
-    s = (
+    s = """Please expand a tip for a topic into a detailed paragraph.
        """Please expand a tip for a topic into a detailed paragraph.
 Topic: staying healthy
 Tip: Regular Exercise
@@ -17,12 +16,7 @@ Topic: writing a blog post
 Tip: structure your content effectively
 Paragraph: A well-structured post is easier to read and more enjoyable. Start with an engaging introduction that hooks the reader and clearly states the purpose of your post. Use headings and subheadings to break up the text and guide readers through your content. Bullet points and numbered lists can make information more digestible. Ensure each paragraph flows logically into the next, and conclude with a summary or call-to-action that encourages reader engagement.
-Topic: """
+Topic: """ + topic + "\nTip: " + tip + "\nParagraph:"
        + topic
        + "\nTip: "
        + tip
        + "\nParagraph:"
    )
    return await generate(s, max_tokens=128, stop="\n\n")
@@ -151,16 +151,14 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "server_process, port = launch_server_cmd(\n",
+    "server_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-8B-Instruct \\\n",
    "    --enable-lora \\\n",
    "    --lora-paths lora0=algoprog/fact-generation-llama-3.1-8b-instruct-lora \\\n",
    "    lora1=Nutanix/Meta-Llama-3.1-8B-Instruct_lora_4_alpha_16 \\\n",
    "    --max-loras-per-batch 2 \\\n",
    "    --log-level warning \\\n",
-    "\"\"\"\n",
+    "\"\"\")\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\", process=server_process)"
   ]
@@ -227,8 +225,7 @@
    "\n",
    "# The `--target-lora-modules` param below is technically not needed, as the server will infer it from lora0 which already has all the target modules specified.\n",
    "# We are adding it here just to demonstrate usage.\n",
-    "server_process, port = launch_server_cmd(\n",
+    "server_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "    python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-8B-Instruct \\\n",
    "    --enable-lora \\\n",
    "    --cuda-graph-max-bs 2 \\\n",
@@ -236,8 +233,7 @@
    "    --max-lora-rank 256\n",
    "    --lora-target-modules all\n",
    "    --log-level warning\n",
-    "    \"\"\"\n",
+    "    \"\"\")\n",
    ")\n",
    "\n",
    "url = f\"http://127.0.0.1:{port}\"\n",
    "wait_for_server(url, process=server_process)"
@@ -435,8 +431,7 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "server_process, port = launch_server_cmd(\n",
+    "server_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "    python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-8B-Instruct \\\n",
    "    --enable-lora \\\n",
    "    --cuda-graph-max-bs 8 \\\n",
@@ -448,8 +443,7 @@
    "        {\"lora_name\":\"lora1\",\"lora_path\":\"algoprog/fact-generation-llama-3.1-8b-instruct-lora\"} \\\n",
    "        lora2=philschmid/code-llama-3-1-8b-text-to-sql-lora\n",
    "    --log-level warning\n",
-    "    \"\"\"\n",
+    "    \"\"\")\n",
    ")\n",
    "\n",
    "\n",
    "url = f\"http://127.0.0.1:{port}\"\n",
@@ -548,16 +542,14 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "server_process, port = launch_server_cmd(\n",
+    "server_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "    python3 -m sglang.launch_server \\\n",
    "    --model-path meta-llama/Meta-Llama-3.1-8B-Instruct \\\n",
    "    --enable-lora \\\n",
    "    --lora-backend csgmv \\\n",
    "    --max-loras-per-batch 16 \\\n",
    "    --lora-paths lora1=path/to/lora1 lora2=path/to/lora2\n",
-    "    \"\"\"\n",
+    "    \"\"\")"
    ")"
   ]
  },
  {
@@ -594,8 +586,7 @@
    "lora2 = \"philschmid/code-llama-3-1-8b-text-to-sql-lora\"\n",
    "\n",
    "\n",
-    "server_process, port = launch_server_cmd(\n",
+    "server_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "    python3 -m sglang.launch_server \\\n",
    "    --model-path meta-llama/Meta-Llama-3.1-8B-Instruct \\\n",
    "    --enable-lora \\\n",
@@ -606,8 +597,7 @@
    "    --max-lora-rank 256 \\\n",
    "    --max-loras-per-batch 2 \\\n",
    "    --max-loaded-loras 4\n",
-    "    \"\"\"\n",
+    "    \"\"\")\n",
    ")\n",
    "\n",
    "url = f\"http://127.0.0.1:{port}\"\n",
    "wait_for_server(url, process=server_process)"
@@ -740,7 +740,6 @@
    "import json\n",
    "from pydantic import BaseModel, Field\n",
    "\n",
    "\n",
    "prompts = [\n",
    "    \"Give me the information of the capital of China in the JSON format.\",\n",
    "    \"Give me the information of the capital of France in the JSON format.\",\n",
@@ -642,7 +642,6 @@
    "import json\n",
    "from pydantic import BaseModel, Field\n",
    "\n",
    "\n",
    "prompts = [\n",
    "    \"Give me the information of the capital of China in the JSON format.\",\n",
    "    \"Give me the information of the capital of France in the JSON format.\",\n",
@@ -117,7 +117,6 @@
   "source": [
    "from sglang import Engine\n",
    "\n",
    "\n",
    "llm = Engine(model_path=model_path, chat_template=chat_template, log_level=\"warning\")"
   ]
  },
@@ -275,12 +275,10 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "embedding_process, port = launch_server_cmd(\n",
+    "embedding_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model-path Alibaba-NLP/gte-Qwen2-1.5B-instruct \\\n",
    "    --host 0.0.0.0 --is-embedding --log-level warning\n",
-    "\"\"\"\n",
+    "\"\"\")\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\", process=embedding_process)"
   ]
@@ -324,12 +322,10 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "reranker_process, port = launch_server_cmd(\n",
+    "reranker_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model-path BAAI/bge-reranker-v2-m3 \\\n",
    "    --host 0.0.0.0 --disable-radix-cache --chunked-prefill-size -1 --attention-backend triton --is-embedding --log-level warning\n",
-    "\"\"\"\n",
+    "\"\"\")\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\", process=reranker_process)"
   ]
@@ -392,12 +388,10 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "score_process, port = launch_server_cmd(\n",
+    "score_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model-path qwen/qwen2.5-0.5b-instruct \\\n",
    "    --host 0.0.0.0 --log-level warning\n",
-    "\"\"\"\n",
+    "\"\"\")\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\", process=score_process)"
   ]
@@ -456,11 +450,9 @@
    "# Note that SGLang now treats embedding models and reward models as the same type of models.\n",
    "# This will be updated in the future.\n",
    "\n",
-    "reward_process, port = launch_server_cmd(\n",
+    "reward_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model-path Skywork/Skywork-Reward-Llama-3.1-8B-v0.2 --host 0.0.0.0 --is-embedding --log-level warning\n",
-    "\"\"\"\n",
+    "\"\"\")\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\", process=reward_process)"
   ]
@@ -575,11 +567,9 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "tokenizer_free_server_process, port = launch_server_cmd(\n",
+    "tokenizer_free_server_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model-path qwen/qwen2.5-0.5b-instruct\n",
-    "\"\"\"\n",
+    "\"\"\")\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\", process=tokenizer_free_server_process)"
   ]
@@ -30,12 +30,10 @@
    "from sglang.test.doc_patch import launch_server_cmd\n",
    "from sglang.utils import wait_for_server, print_highlight, terminate_process\n",
    "\n",
-    "embedding_process, port = launch_server_cmd(\n",
+    "embedding_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model-path Alibaba-NLP/gte-Qwen2-1.5B-instruct \\\n",
    "    --host 0.0.0.0 --is-embedding --log-level warning\n",
-    "\"\"\"\n",
+    "\"\"\")\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\", process=embedding_process)"
   ]
@@ -33,11 +33,9 @@
    "from sglang.test.doc_patch import launch_server_cmd\n",
    "from sglang.utils import wait_for_server, print_highlight, terminate_process\n",
    "\n",
-    "vision_process, port = launch_server_cmd(\n",
+    "vision_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model-path Qwen/Qwen2.5-VL-7B-Instruct --log-level warning\n",
-    "\"\"\"\n",
+    "\"\"\")\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\", process=vision_process)"
   ]
@@ -31,12 +31,10 @@
    "# This is equivalent to running the following command in your terminal\n",
    "# python3 -m sglang.launch_server --model-path qwen/qwen2.5-0.5b-instruct --host 0.0.0.0\n",
    "\n",
-    "server_process, port = launch_server_cmd(\n",
+    "server_process, port = launch_server_cmd(\"\"\"\n",
    "    \"\"\"\n",
    "python3 -m sglang.launch_server --model-path qwen/qwen2.5-0.5b-instruct \\\n",
    " --host 0.0.0.0 --log-level warning\n",
-    "\"\"\"\n",
+    "\"\"\")\n",
    ")\n",
    "\n",
    "wait_for_server(f\"http://localhost:{port}\", process=server_process)"
   ]
@@ -107,7 +107,6 @@
    "from sglang.lang.interpreter import ProgramState\n",
    "from parea import Parea, trace\n",
    "\n",
    "\n",
    "load_dotenv()\n",
    "\n",
    "os.environ[\"TOKENIZERS_PARALLELISM\"] = \"false\"\n",
@@ -237,7 +236,6 @@
    "    percent_target_supported_by_context_factory,\n",
    ")\n",
    "\n",
    "\n",
    "context_relevancy_eval = context_query_relevancy_factory()\n",
    "percent_target_supported_by_context = percent_target_supported_by_context_factory()\n",
    "\n",
@@ -263,7 +261,6 @@
    "from parea.evals.general import answer_matches_target_llm_grader_factory\n",
    "from parea.evals.rag import answer_context_faithfulness_statement_level_factory\n",
    "\n",
    "\n",
    "answer_context_faithfulness = answer_context_faithfulness_statement_level_factory()\n",
    "answer_matches_target_llm_grader = answer_matches_target_llm_grader_factory()\n",
    "\n",
@@ -18,6 +18,7 @@ llm = Engine(
    tensor_parallel_size=8,
 )
 """
 import dataclasses
 from argparse import ArgumentParser
 from pathlib import Path
@@ -21,6 +21,7 @@ llm = Engine(
    tensor_parallel_size=8,
 )
 """
 import dataclasses
 import os
 import shutil
@@ -36,7 +36,8 @@ struct dtype_trait {};
  }                                                                 \
  static_assert(true)
-SGL_REGISTER_DTYPE_TRAIT(fp32_t, fp32x2_t, SGL_REGISTER_TYPE_END;  //
+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);
@@ -48,15 +49,15 @@ 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_DTYPE_TRAIT(
-                         SGL_REGISTER_FROM_FUNCTION(fp16x2_t, __half22float2);
+    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_DTYPE_TRAIT(
-                         SGL_REGISTER_FROM_FUNCTION(fp32x2_t, __float22half2_rn););
+    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_DTYPE_TRAIT(
-                         SGL_REGISTER_FROM_FUNCTION(fp32x2_t, __float22bfloat162_rn););
+    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
@@ -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(
+        print(f"""
            f"""
 {delimiter}
 \033[1mSGLang Diffusion WebUI available at:\033[0m \033[1;4;92m{url}\033[0m
 {delimiter}
-"""
+""")
        )
        demo.block_thread()
@@ -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 (
@@ -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,
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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,
@@ -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.
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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,
 )
@@ -4,6 +4,7 @@
 """
 Synchronous pipeline executor implementation.
 """
 from typing import List
 from sglang.multimodal_gen.runtime.pipelines_core.executors.pipeline_executor import (
@@ -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
@@ -4,6 +4,7 @@
 """
 Input validation stage for diffusion pipelines.
 """
 import numpy as np
 import torch
 import torchvision.transforms.functional as TF
@@ -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
@@ -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
@@ -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
@@ -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
@@ -3,6 +3,7 @@
 """
 Common generate cli test, one test for image and video each
 """
 import dataclasses
 import os
 import shlex
@@ -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
@@ -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
@@ -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,
@@ -15,6 +15,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """EXAONE model configuration"""
 from typing import Any, Dict
 from transformers.configuration_utils import PretrainedConfig
@@ -14,7 +14,6 @@
 # limitations under the License.
 """Falcon-H1 model configuration"""
 from transformers.configuration_utils import PretrainedConfig
 from transformers.utils import logging
@@ -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"))
@@ -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
@@ -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
@@ -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]
@@ -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
@@ -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_proj_states, k_proj_states, v_proj_states = mixed_qkv
-        (q_conv_weights, k_conv_weights, v_conv_weights) = layer.conv_weights
+        q_conv_weights, k_conv_weights, v_conv_weights = layer.conv_weights
-        (q_conv_bias, k_conv_bias, v_conv_bias) = layer.bias
+        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_proj_states, k_proj_states, v_proj_states = mixed_qkv
-        (q_conv_weights, k_conv_weights, v_conv_weights) = layer.conv_weights
+        q_conv_weights, k_conv_weights, v_conv_weights = layer.conv_weights
-        (q_conv_bias, k_conv_bias, v_conv_bias) = layer.bias
+        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
@@ -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
@@ -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,
@@ -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,
@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 """Cutlass W4A8 MoE kernel."""
 from typing import Optional
 import torch
@@ -63,11 +63,9 @@ class ModelSlimMoEMethod(FusedMoEMethodBase):
            logger.info_once("Using ModelSlimW8A8Int8MoE")
            return ModelSlimW8A8Int8MoE(quant_config)
        else:
-            logger.warning(
+            logger.warning(f"Unsupported FusedMoe modelslim scheme: \
                f"Unsupported FusedMoe modelslim scheme: \
                    {quant_config.quant_description.get(prefix_in_quant_config.strip())} \
-                    in layer: {prefix}"
+                    in layer: {prefix}")
            )
            return None
@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """Radix attention."""
 from __future__ import annotations
 from enum import Enum
@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """Radix linear attention."""
 from __future__ import annotations
 from typing import TYPE_CHECKING, Optional, Tuple, Union
@@ -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()
        )
@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """A tensor parallel worker."""
 from __future__ import annotations
 import logging
@@ -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,
        )
@@ -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,
 )
@@ -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,
        )
@@ -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,
        )
@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """Utilities for Prometheus Metrics Collection."""
 import dataclasses
 import logging
 import os
@@ -12,6 +12,7 @@
 # limitations under the License.
 # ==============================================================================
 """Utilities for Prometheus Metrics."""
 import math
 from typing import List
@@ -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
@@ -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(
+        raise RuntimeError(f"""The file size different is more than 1%:
            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)
@@ -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
@@ -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
@@ -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
@@ -13,6 +13,7 @@
 # ==============================================================================
 """Inference-only DeepSeek NextN Speculative Decoding."""
 import logging
 from typing import Iterable, Optional, Tuple
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -13,6 +13,7 @@
 # ==============================================================================
 """Inference-only OPT model compatible with HuggingFace weights."""
 import logging
 from collections.abc import Iterable
 from typing import Optional, Union
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -13,6 +13,7 @@
 # ==============================================================================
 """Inference-only Qwen3_5 MTP model."""
 import logging
 from typing import Iterable, Optional, Tuple
@@ -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)
+        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, 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)
@@ -13,6 +13,7 @@
 # ==============================================================================
 """Inference-only Qwen3Next MTP Speculative Decoding."""
 import logging
 from typing import Iterable, Optional, Tuple
@@ -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
@@ -13,6 +13,7 @@
 # limitations under the License.
 # ==============================================================================
 """Inference-only Qwen3-VL model compatible with HuggingFace weights."""
 import logging
 import math
 import re
@@ -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
@@ -21,6 +21,7 @@
 # limitations under the License.
 # Adapted from https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/starcoder2.py
 """PyTorch Starcoder2 model."""
 from collections.abc import Iterable
 from typing import Optional, Tuple
@@ -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
--- a/Show More
+++ b/Show More