[Bugfix] Fix fp8 tests for triton_unified_attention for Triton 3.3 (#18013)
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com> Co-authored-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
This commit is contained in:
co-authored by
Lucas Wilkinson
parent
26d0419309
commit
e6b8e65d2d
@@ -268,6 +268,10 @@ def unified_attention(
|
||||
assert causal, "Only causal attention is supported"
|
||||
assert q_descale is None, "Q scales not supported"
|
||||
|
||||
block_size = v.shape[1]
|
||||
assert q.element_size() >= 2 or block_size >= 32, \
|
||||
"Block size must be at least 32 for fp8"
|
||||
|
||||
use_alibi_slopes = alibi_slopes is not None
|
||||
|
||||
block_size = v.shape[1]
|
||||
|
||||
Reference in New Issue
Block a user