diff --git a/tests/data/test_valid_seqlens_devices.py b/tests/data/test_valid_seqlens_devices.py
new file mode 100644
index 0000000000..0f2719496c
--- /dev/null
+++ b/tests/data/test_valid_seqlens_devices.py
@@ -0,0 +1,73 @@
+# Copyright 2026 Bytedance Ltd. and/or its affiliates
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
+"""Tail-segment extraction must agree on CPU and Ascend NPU."""
+
+import json
+import os
+from pathlib import Path
+
+import pytest
+import torch
+
+from scripts.ci.accelerator.launch_plan import LaunchSpec
+from veomni.utils.device import IS_NPU_AVAILABLE
+from veomni.utils.seqlen_pos_transform_utils import valid_seqlens_from_cu_seqlens
+
+
+_CASES = [
+    ([0, 4, 7, 8, 9, 10], None, [4, 3]),
+    ([0, 4, 5, 6, 7, 8], 3, [4, 1]),
+    ([0, 4, 5], 0, [4, 1]),
+    ([0, 1, 3], None, [1, 2]),
+    ([0, 1, 2], None, []),
+    ([0], None, []),
+]
+
+
+def _npu_test_eligible():
+    """Honor the CI lease; on NPU CI leave device availability to the test assertion."""
+    capabilities_file = os.environ.get("VEOMNI_CI_CAPABILITIES")
+    if capabilities_file:
+        return json.loads(Path(capabilities_file).read_text())["backend"] == "npu"
+    return IS_NPU_AVAILABLE and torch.npu.is_available()
+
+
+@pytest.mark.launch_plan(provider=LaunchSpec(device="cpu"))
+@pytest.mark.parametrize(("backend", "expected"), [("cpu", False), ("mlu", False), ("npu", True)])
+def test_npu_eligibility_respects_ci_backend(monkeypatch, tmp_path, backend, expected):
+    capabilities = tmp_path / "capabilities.json"
+    capabilities.write_text(json.dumps({"backend": backend, "devices": []}))
+    monkeypatch.setenv("VEOMNI_CI_CAPABILITIES", str(capabilities))
+    assert _npu_test_eligible() is expected
+
+
+@pytest.mark.launch_plan(provider=LaunchSpec(device="cpu"))
+@pytest.mark.parametrize("cu,tail,expected", _CASES)
+def test_valid_seqlens_cpu(cu, tail, expected):
+    value = torch.tensor(cu, dtype=torch.int32)
+    result = valid_seqlens_from_cu_seqlens(value, tail_padding_length=tail)
+    assert result.tolist() == expected
+    assert result.dtype == torch.int32
+
+
+@pytest.mark.launch_plan(provider=LaunchSpec(device="accelerator"))
+@pytest.mark.skipif(not _npu_test_eligible(), reason="requires an Ascend NPU")
+@pytest.mark.parametrize("cu,tail,expected", _CASES)
+def test_valid_seqlens_ascend(cu, tail, expected):
+    assert IS_NPU_AVAILABLE and torch.npu.is_available(), "Ascend CI must provide an available NPU"
+    value = torch.tensor(cu, dtype=torch.int32, device="npu")
+    result = valid_seqlens_from_cu_seqlens(value, tail_padding_length=tail)
+    assert result.cpu().tolist() == expected
+    assert result.dtype == torch.int32
diff --git a/veomni/utils/seqlen_pos_transform_utils.py b/veomni/utils/seqlen_pos_transform_utils.py
index 2d6adb58ae..7032528767 100644
--- a/veomni/utils/seqlen_pos_transform_utils.py
+++ b/veomni/utils/seqlen_pos_transform_utils.py
@@ -157,5 +157,5 @@ def valid_seqlens_from_cu_seqlens(cu_seqlens: torch.Tensor, tail_padding_length:
         valid_end = torch.clamp(cu_seqlens[-1:] - tail_padding_length, min=0)
         return diff[cu_seqlens[1:] <= valid_end]
 
-    pad = int((torch.flip(diff == 1, (0,)).cumprod(0)).sum().item())
+    pad = int((torch.flip(diff == 1, (0,)).int().cumprod(0)).sum().item())
     return diff[:-pad] if pad else diff