Operator Support#
This page lists the PyTorch operators currently supported by the Samsung Exynos backend.
Operator |
Quantization |
Constraints |
|---|---|---|
add |
static int8 |
|
avg_pool2d |
static int8 |
ceil_mode=False, divisor_override=pooling_region |
batch_norm |
static int8 |
|
bmm |
static int8 |
|
cat |
static int8 |
at most 1 constant tensor |
clamp |
static int8 |
|
constant_pad_nd |
static int8 |
padding_value=0.0 only |
conv2d |
static int8 |
constant weights |
dequantize_per_channel |
||
dequantize_per_tensor |
||
div |
static int8 |
|
embedding |
static int8 |
|
expand_copy |
expanding at most one axis, new dimensions must be size 1 |
|
gelu |
static int8 |
|
getitem |
||
hardsigmoid |
static int8 |
|
hardswish |
static int8 |
|
hardtanh |
static int8 |
|
layer_norm |
static int8 |
norm at last axis only |
leaky_relu |
static int8 |
|
linear |
static int8 |
constant weights |
log_softmax |
static int8 |
|
max_pool2d |
static int8 |
ceil_mode=False, indices not supported |
maximum |
||
mean_dim |
static int8 |
|
minimum |
||
mul |
static int8 |
|
permute |
static int8 |
|
pixel_shuffle |
||
quantize_per_channel |
||
quantize_per_tensor |
||
relu |
static int8 |
|
reshape |
static int8 |
|
rsqrt |
static int8 |
|
select |
static int8 |
|
slice_copy |
static int8 |
|
softmax |
static int8 |
|
sqrt |
static int8 |
|
squeeze |
static int8 |
|
sub |
static int8 |
|
to_copy |
memory_format=contiguous only |
|
unsqueeze |
static int8 |
|
upsample_bilinear2d |
static int8 |
|
upsample_nearest2d |
static int8 |
