[Attention][FlashInfer] Enable FP8 FlashInfer (TRTLLM) MLA decode (#24705)

Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com>
This commit is contained in:
Matthew Bonanni
2025-09-12 15:45:53 -06:00
committed by GitHub
parent c89ed8de43
commit 7ba32aa60b
8 changed files with 23 additions and 10 deletions
@@ -178,6 +178,7 @@ class MockAttentionLayer:
self._k_scale = torch.tensor(1.0, device=device)
self._v_scale = torch.tensor(1.0, device=device)
# Add float versions for flashinfer
self._q_scale_float = 1.0
self._k_scale_float = 1.0
self._v_scale_float = 1.0
+2
View File
@@ -33,10 +33,12 @@ def test_ragged_paged_attention():
)
class FakeAttentionLayer:
_q_scale_float: float
_k_scale_float: float
_v_scale_float: float
layer = FakeAttentionLayer()
layer._q_scale_float = 1.0
layer._k_scale_float = 1.0
layer._v_scale_float = 1.0