[Lint] Add `python/sglang` to ruff F401 checks and remove unused imports in files (#11685)

python/sglang/srt/function_call/glm4_moe_detector.py

@@ -6,11 +6,7 @@ from typing import List
 
 from sglang.srt.entrypoints.openai.protocol import Tool
 from sglang.srt.function_call.base_format_detector import BaseFormatDetector
-from sglang.srt.function_call.core_types import (
-    StreamingParseResult,
-    StructureInfo,
-    _GetInfoFunc,
-)
+from sglang.srt.function_call.core_types import StreamingParseResult, _GetInfoFunc
 from sglang.srt.function_call.ebnf_composer import EBNFComposer
 
 logger = logging.getLogger(__name__)

python/sglang/srt/function_call/json_array_parser.py

-import json
-import re
 from typing import List
 
 from sglang.srt.entrypoints.openai.protocol import Tool

python/sglang/srt/function_call/utils.py

-import json
 from json import JSONDecodeError, JSONDecoder
 from json.decoder import WHITESPACE
 from typing import Any, List, Literal, Optional, Tuple, Union

python/sglang/srt/grpc/compile_proto.py

@@ -70,7 +70,7 @@ def compile_proto(proto_file: Path, output_dir: Path, verbose: bool = True) -> b
 
     # Check if grpc_tools is available
     try:
-        import grpc_tools.protoc
+        import grpc_tools.protoc  # noqa: F401
     except ImportError:
         print("Error: grpcio-tools not installed")
         print(

python/sglang/srt/grpc/grpc_request_manager.py

@@ -27,7 +27,6 @@ from sglang.srt.managers.io_struct import (
     TokenizedEmbeddingReqInput,
     TokenizedGenerateReqInput,
 )
-from sglang.srt.managers.scheduler import is_health_check_generate_req
 from sglang.srt.server_args import PortArgs, ServerArgs
 from sglang.srt.utils import get_zmq_socket, kill_process_tree
 from sglang.utils import get_exception_traceback

python/sglang/srt/layers/activation.py

@@ -380,4 +380,7 @@ if not (
     logger.info(
         "sgl-kernel is not available on Non-NV, Non-AMD platforms or Non-AMX CPUs. Fallback to other kernel libraries."
     )
-    from vllm.model_executor.layers.activation import GeluAndMul, SiluAndMul
+    from vllm.model_executor.layers.activation import (  # noqa: F401
+        GeluAndMul,
+        SiluAndMul,
+    )

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

@@ -20,7 +20,6 @@ if TYPE_CHECKING:
     from sglang.srt.layers.radix_attention import RadixAttention
     from sglang.srt.model_executor.model_runner import ModelRunner
 
-import os
 
 import numpy as np
 

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

 from __future__ import annotations
 
 from abc import ABC, abstractmethod
-from typing import TYPE_CHECKING, Optional, Union
+from typing import TYPE_CHECKING, Optional
 
 import torch
 

python/sglang/srt/layers/attention/fla/chunk.py

@@ -2,7 +2,6 @@
 # -*- coding: utf-8 -*-
 # Copyright (c) 2023-2025, Songlin Yang, Yu Zhang
 
-import warnings
 from typing import Optional
 
 import torch

python/sglang/srt/layers/attention/fla/chunk_o.py

@@ -2,7 +2,7 @@
 # -*- coding: utf-8 -*-
 # Copyright (c) 2023-2025, Songlin Yang, Yu Zhang
 
-from typing import Optional, Tuple
+from typing import Optional
 
 import torch
 import triton

python/sglang/srt/layers/attention/fla/index.py

@@ -3,9 +3,7 @@
 # Copyright (c) 2023-2025, Songlin Yang, Yu Zhang
 
 import torch
-import torch.nn.functional as F
 import triton
-import triton.language as tl
 
 from sglang.srt.layers.attention.fla.utils import tensor_cache
 

python/sglang/srt/layers/attention/fla/layernorm_gated.py

@@ -5,7 +5,6 @@
 # This backward pass is faster for dimensions up to 8k, but after that it's much slower due to register spilling.
 # The models we train have hidden dim up to 8k anyway (e.g. Llama 70B), so this is fine.
 
-import math
 
 import torch
 import torch.nn.functional as F

python/sglang/srt/layers/attention/fla/wy_fast.py

@@ -9,8 +9,6 @@ import triton
 import triton.language as tl
 
 from sglang.srt.layers.attention.fla.index import prepare_chunk_indices
-from sglang.srt.layers.attention.fla.op import safe_exp
-from sglang.srt.layers.attention.fla.utils import check_shared_mem
 
 
 @triton.heuristics({"IS_VARLEN": lambda args: args["cu_seqlens"] is not None})

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

@@ -50,7 +50,6 @@ if is_flashinfer_available():
         fast_decode_plan,
     )
     from flashinfer.cascade import merge_state
-    from flashinfer.decode import _get_range_buf, get_seq_lens
 
 
 class WrapperDispatch(Enum):

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

-from typing import Optional, Union
+from typing import Optional
 
 import torch
 

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

-from dataclasses import astuple, dataclass
-from functools import lru_cache
 from typing import Optional, Union
 
 import torch
-import torch.nn.functional as F
 
 from sglang.srt.layers.attention.base_attn_backend import AttentionBackend
 from sglang.srt.layers.attention.fla.chunk import chunk_gated_delta_rule

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

@@ -14,7 +14,7 @@ if TYPE_CHECKING:
 
 class IntelAMXAttnBackend(AttentionBackend):
     def __init__(self, model_runner: ModelRunner):
-        import sgl_kernel
+        import sgl_kernel  # noqa: F401
 
         super().__init__()
         self.forward_metadata = None

python/sglang/srt/layers/attention/mamba/causal_conv1d_triton.py

@@ -4,7 +4,6 @@
 
 from typing import List, Optional, Union
 
-import numpy as np
 import torch
 import triton
 import triton.language as tl

python/sglang/srt/layers/attention/mamba/ops/ssd_combined.py

@@ -10,7 +10,6 @@
 
 import torch
 import triton
-import triton.language as tl
 from einops import rearrange
 from packaging import version
 

python/sglang/srt/layers/attention/npu_ops/mla_preprocess.py

@@ -13,7 +13,7 @@ def is_mla_preprocess_enabled() -> bool:
 
 
 if is_mla_preprocess_enabled():
-    import sgl_kernel_npu
+    import sgl_kernel_npu  # noqa: F401
     import torch_npu
 
     torch.npu.config.allow_internal_format = True