[ROCm][CI] Fix ModernBERT token classification test numerical accuracy on ROCm (#31820)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>

tests/models/language/pooling/conftest.py

+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""Pytest configuration for vLLM language generation tests."""
+
+import warnings
+
+import torch
+
+from vllm.platforms import current_platform
+
+
+def pytest_sessionstart(session):
+    """Configure ROCm-specific settings before test session starts."""
+    if not current_platform.is_rocm():
+        return
+
+    # Disable Flash/MemEfficient SDP on ROCm to avoid HF Transformers
+    # accuracy issues: https://github.com/vllm-project/vllm/issues/30167
+    # TODO: Remove once ROCm SDP accuracy issues are resolved on HuggingFace
+    torch.backends.cuda.enable_flash_sdp(False)
+    torch.backends.cuda.enable_mem_efficient_sdp(False)
+    torch.backends.cuda.enable_math_sdp(True)
+    torch.set_float32_matmul_precision("high")
+    warnings.warn(
+        "ROCm: Disabled flash_sdp and mem_efficient_sdp, enabled math_sdp "
+        "to avoid HuggingFace Transformers accuracy issues",
+        UserWarning,
+        stacklevel=1,
+    )