[Attention] Integrate FlashAttention 4 SM100 headdim 256 support (#42669)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
This commit is contained in:
co-authored by
gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
parent
8112b6c997
commit
bf2b45b5d6
@@ -163,7 +163,7 @@ class TestGetMLAPrefillBackend:
|
||||
class TestAutoSelectMLAPrefillBackend:
|
||||
"""Tests for fallback and error paths in auto-selection."""
|
||||
|
||||
def test_blackwell_glm_dimensions_fall_back_to_trtllm(self):
|
||||
def test_blackwell_glm_dimensions_use_trtllm(self):
|
||||
capability = DeviceCapability(major=10, minor=0)
|
||||
selector_config = MLAPrefillSelectorConfig(
|
||||
dtype=torch.bfloat16,
|
||||
@@ -184,11 +184,6 @@ class TestAutoSelectMLAPrefillBackend:
|
||||
with (
|
||||
patch("vllm.platforms.current_platform") as mock_platform,
|
||||
patch.object(flash_attn_cls, "is_available", return_value=True),
|
||||
patch(
|
||||
"vllm.v1.attention.backends.mla.prefill.flash_attn."
|
||||
"get_flash_attn_version",
|
||||
return_value=4,
|
||||
),
|
||||
patch.object(trtllm_cls, "validate_configuration", return_value=[]),
|
||||
):
|
||||
# Force the non-ROCm priority on the Blackwell.
|
||||
@@ -300,7 +295,12 @@ class TestROCmAiterFAPrefillSelection:
|
||||
with patch("vllm.platforms.current_platform") as mock_platform:
|
||||
mock_platform.is_rocm.return_value = True
|
||||
priorities = _get_mla_prefill_backend_priorities(
|
||||
DeviceCapability(major=9, minor=5)
|
||||
DeviceCapability(major=9, minor=5),
|
||||
MLADimensions(
|
||||
qk_nope_head_dim=128,
|
||||
qk_rope_head_dim=64,
|
||||
v_head_dim=128,
|
||||
),
|
||||
)
|
||||
|
||||
assert priorities == [
|
||||
|
||||
Reference in New Issue
Block a user