[ROCm] Fix AITER_UNIFIED_ATTN Dispatching After AITER Bump (#46780)
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Signed-off-by: Rohan138 <rohanpotdar138@gmail.com> Co-authored-by: Rohan138 <rohanpotdar138@gmail.com>
This commit is contained in:
co-authored by
Rohan138
parent
bf292b5f6b
commit
d980a3cc6e
@@ -30,7 +30,8 @@ NUM_Q_HEADS = 8
|
||||
NUM_KV_HEADS = 8
|
||||
HEAD_SIZES = [128, 256]
|
||||
BLOCK_SIZES = [16, 64]
|
||||
DTYPES = [torch.bfloat16, torch.float16]
|
||||
# TODO: re-add torch.float16 once AITER reenables fp16 unified attention.
|
||||
DTYPES = [torch.bfloat16]
|
||||
FP8_DTYPE = current_platform.fp8_dtype()
|
||||
|
||||
# (query_len, kv_len) per sequence
|
||||
|
||||
Reference in New Issue
Block a user