[Hardware][AMD][CI][Bugfix] Fix Kernels Attention Cache test (#32904)
Signed-off-by: Matthew Wong <Matthew.Wong2@amd.com>
This commit is contained in:
@@ -242,7 +242,7 @@ def test_reshape_and_cache_flash(
|
|||||||
value_cache_compact = permute_and_compact(value_cache)
|
value_cache_compact = permute_and_compact(value_cache)
|
||||||
|
|
||||||
def convert_fp8_local(output, input, scale, kv_dtype):
|
def convert_fp8_local(output, input, scale, kv_dtype):
|
||||||
fp8_input = input.view(torch.float8_e4m3fn)
|
fp8_input = input.view(current_platform.fp8_dtype())
|
||||||
if scale.numel() == 1: # per-tensor
|
if scale.numel() == 1: # per-tensor
|
||||||
result = scaled_dequantize(
|
result = scaled_dequantize(
|
||||||
fp8_input.flatten(0, 2), scale, group_shape=None, out_dtype=output.dtype
|
fp8_input.flatten(0, 2), scale, group_shape=None, out_dtype=output.dtype
|
||||||
|
|||||||
Reference in New Issue
Block a user