[ROCm] Remove stale SDPA and skinny GEMM workarounds (#50907)

Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com>
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: OpenAI Codex <codex@openai.com>

tests/entrypoints/pooling/basic/test_encode.py

@@ -7,7 +7,6 @@ import pytest
 
 from vllm import LLM, PoolingParams
 from vllm.distributed import cleanup_dist_env_and_memory
-from vllm.platforms import current_platform
 
 MODEL_NAME = "intfloat/multilingual-e5-small"
 
@@ -30,12 +29,6 @@ TOKEN_IDS = [
 
 @pytest.fixture(scope="module")
 def llm():
-    # ROCm: Use FLEX_ATTENTION backend as it's the only attention backend
-    # that supports encoder-only models on ROCm.
-    attention_config = None
-    if current_platform.is_rocm():
-        attention_config = {"backend": "FLEX_ATTENTION"}
-
     # pytest caches the fixture so we use weakref.proxy to
     # enable garbage collection
     llm = LLM(
@@ -45,7 +38,6 @@ def llm():
         gpu_memory_utilization=0.75,
         enforce_eager=True,
         seed=0,
-        attention_config=attention_config,
     )
 
     yield weakref.proxy(llm)