diff --git a/vllm/model_executor/specialized_models/deepseek_v3_2_nvfp4/layer.py b/vllm/model_executor/specialized_models/deepseek_v3_2_nvfp4/layer.py
index eef6521815d..cbdb0adcfc7 100644
--- a/vllm/model_executor/specialized_models/deepseek_v3_2_nvfp4/layer.py
+++ b/vllm/model_executor/specialized_models/deepseek_v3_2_nvfp4/layer.py
@@ -65,7 +65,7 @@ def dsa(
         output.zero_()
         return output
 
-    num_actual_toks = mla_attn_metadata.num_actual_tokens  # noqa
+    num_actual_toks = mla_attn_metadata.num_actual_tokens  # type: ignore[attr-defined]
     if num_actual_toks == 0:
         output.zero_()
         return output
@@ -77,7 +77,7 @@ def dsa(
     mla_k_scale = None
     idx_meta = attn_metadata.get(attn.indexer_k_cache.prefix)
     if idx_meta is not None:
-        slot_mapping = idx_meta.slot_mapping  # noqa
+        slot_mapping = idx_meta.slot_mapping  # type: ignore[attr-defined]
         indexer_k_cache = attn.indexer_k_cache.kv_cache
         mla_kv_cache = attn.mla_attn.kv_cache
         mla_k_scale = attn.mla_attn._k_scale
diff --git a/vllm/model_executor/specialized_models/deepseek_v3_2_nvfp4/sparse_indexer.py b/vllm/model_executor/specialized_models/deepseek_v3_2_nvfp4/sparse_indexer.py
index 582e8db2dbc..51a223290d4 100644
--- a/vllm/model_executor/specialized_models/deepseek_v3_2_nvfp4/sparse_indexer.py
+++ b/vllm/model_executor/specialized_models/deepseek_v3_2_nvfp4/sparse_indexer.py
@@ -5,6 +5,7 @@
 import torch
 
 import vllm.envs as envs
+from vllm import _custom_ops as ops
 from vllm.forward_context import get_forward_context
 from vllm.logger import init_logger
 from vllm.platforms import current_platform
@@ -15,11 +16,6 @@ from vllm.v1.attention.backends.mla.indexer import (
 from vllm.v1.attention.ops.common import pack_seq_triton, unpack_seq_triton
 from vllm.v1.worker.workspace import current_workspace_manager
 
-if current_platform.is_cuda_alike():
-    from vllm import _custom_ops as ops
-elif current_platform.is_xpu():
-    from vllm._xpu_ops import xpu_ops as ops
-
 logger = init_logger(__name__)
 
 RADIX_TOPK_WORKSPACE_SIZE = 1024 * 1024