Merge tag 'v0.7.2' into v0.7.2-dev

tests/kernels/test_moe.py

+# SPDX-License-Identifier: Apache-2.0
 """Tests for the MOE layers.
 
 Run `pytest tests/kernels/test_moe.py`.

tests/kernels/test_pos_encoding.py

+# SPDX-License-Identifier: Apache-2.0
+
 from itertools import accumulate, product
 from typing import Dict, List, Optional
 

tests/kernels/test_prefix_prefill.py

+# SPDX-License-Identifier: Apache-2.0
+
 import math
 import random
 import time

tests/kernels/test_rocm_attention_selector.py

+# SPDX-License-Identifier: Apache-2.0
+
+from unittest.mock import patch
+
+import pytest
+import torch
+
+from tests.kernels.utils import override_backend_env_variable
+from vllm.attention.selector import _cached_get_attn_backend, get_attn_backend
+from vllm.platforms.rocm import RocmPlatform
+
+
+@pytest.fixture(autouse=True)
+def clear_cache():
+    """Clear lru cache to ensure each test case runs without caching.
+    """
+    _cached_get_attn_backend.cache_clear()
+
+
+def test_selector(monkeypatch):
+    """Test that the attention selector for ROCm.
+    """
+    override_backend_env_variable(monkeypatch, "ROCM_FLASH")
+
+    with patch("vllm.attention.selector.current_platform", RocmPlatform()):
+        backend = get_attn_backend(16, torch.float16, torch.float16, 16, False)
+        assert backend.get_name() == "ROCM_FLASH"
+        # mla test for deepseek related
+        backend = get_attn_backend(576, torch.bfloat16, "auto", 16, False,
+                                   False, True)
+        assert backend.get_name() == "TRITON_MLA"

tests/kernels/test_rotary_embedding.py

+# SPDX-License-Identifier: Apache-2.0
 """
 Tests for miscellaneous utilities
 """

tests/kernels/test_triton_decode_attention.py

+# SPDX-License-Identifier: Apache-2.0
+
 import pytest
 import torch
 

tests/kernels/test_triton_scaled_mm.py

+# SPDX-License-Identifier: Apache-2.0
 """Tests for the triton_scaled_mm kernel
 
 Run `pytest tests/kernels/test_triton_scaled_mm.py`.

tests/kernels/test_utils.py

+# SPDX-License-Identifier: Apache-2.0
 """
 Tests for miscellaneous utilities
 """

tests/kernels/untest_aqlm.py

+# SPDX-License-Identifier: Apache-2.0
+
 import torch
 
 from tests.kernels.utils import opcheck

tests/kernels/untest_awq.py

+# SPDX-License-Identifier: Apache-2.0
+
 import os
 
 import pytest

tests/kernels/untest_causal_conv1d.py

+# SPDX-License-Identifier: Apache-2.0
+
 from typing import Optional
 
 import pytest

tests/kernels/untest_flashinfer.py

+# SPDX-License-Identifier: Apache-2.0
+
 from typing import List, Optional, Tuple
 
 import flashinfer

tests/kernels/untest_fp8_quant.py

+# SPDX-License-Identifier: Apache-2.0
+
 import pytest
 import torch
 

tests/kernels/untest_ggml.py

+# SPDX-License-Identifier: Apache-2.0
+
 import gguf
 import pytest
 import torch

tests/kernels/untest_gptq.py

+# SPDX-License-Identifier: Apache-2.0
+
 import torch
 
 from tests.kernels.utils import opcheck

tests/kernels/untest_mamba_ssm.py

+# SPDX-License-Identifier: Apache-2.0
+
 import pytest
 import torch
 import torch.nn.functional as F

tests/kernels/untest_marlin_gemm.py

+# SPDX-License-Identifier: Apache-2.0
 """Tests for the marlin kernel.
 
 Run `pytest tests/kernels/marlin/test_marlin_gemm.py`.

tests/kernels/untest_permute_cols.py

+# SPDX-License-Identifier: Apache-2.0
+
 import pytest
 import torch
 

tests/kernels/utils.py

+# SPDX-License-Identifier: Apache-2.0
 """Kernel test utils"""
 
 import itertools

tests/kv_transfer/disagg_test.py

+# SPDX-License-Identifier: Apache-2.0
+
 import os
 import subprocess
 import sys