diff options
author | Eric Kunze <eric.kunze@arm.com> | 2020-10-30 14:26:40 -0700 |
---|---|---|
committer | Eric Kunze <eric.kunze@arm.com> | 2020-11-03 11:21:27 -0800 |
commit | b0b4bf083b9e73e8f3e96a02a612fac53d3ad651 (patch) | |
tree | 9b0b10239f1193598a427d55b9bfd47bae8a1611 /chapters | |
parent | 408f7c1a2986896bb1c7b2c6515f0dbe079d05c9 (diff) | |
download | specification-b0b4bf083b9e73e8f3e96a02a612fac53d3ad651.tar.gz |
Change weights to input for training profile
Weights remain as an attribute for inference profile devices.
Change-Id: I1cdbbd2ccd95fc1b7afc0f05a1425fb53560b432
Signed-off-by: Eric Kunze <eric.kunze@arm.com>
Diffstat (limited to 'chapters')
-rw-r--r-- | chapters/ewise_binary.adoc | 2 | ||||
-rw-r--r-- | chapters/tensor_ops.adoc | 16 | ||||
-rw-r--r-- | chapters/type_conversion.adoc | 6 |
3 files changed, 12 insertions, 12 deletions
diff --git a/chapters/ewise_binary.adoc b/chapters/ewise_binary.adoc index 7292962..a9787b3 100644 --- a/chapters/ewise_binary.adoc +++ b/chapters/ewise_binary.adoc @@ -479,7 +479,7 @@ Elementwise multiplication (Hadamard product) of input tensor 0 and input tensor |Input|in_t*|input1|shape1|Input tensor |Input|in_t*|input2|shape2|Input tensor with the same rank as Input 0 -|Attribute|uint6_t|shift|-|Result right shift (int32 data type only) +|Input (MT profile) Attribute (BI/MI profiles)|uint6_t|shift|-|Result right shift (int32 data type only) |Output|out_t*|output|shape|Output tensor with broadcast shape if necessary |=== diff --git a/chapters/tensor_ops.adoc b/chapters/tensor_ops.adoc index b8e053c..9bb6496 100644 --- a/chapters/tensor_ops.adoc +++ b/chapters/tensor_ops.adoc @@ -132,8 +132,8 @@ Performs a 2D convolution over the given tensor input, using the weight tensor. |Argument|Type|Name|Shape|Description |Input|in_t*|input|[N,IH,IW,IC]|Input tensor -|Attribute|weight_t*|weight|[OC,KH,KW,IC]|Weight kernel size KH x KW -|Attribute|acc_t*|bias|[OC]|Per output channel bias data. +|Input (MT profile) Attribute (BI/MI profiles)|weight_t*|weight|[OC,KH,KW,IC]|Weight kernel size KH x KW +|Input (MT profile) Attribute (BI/MI profiles)|acc_t*|bias|[OC]|Per output channel bias data. |Attribute|int*|pad|[4]|[pad_top, pad_bottom, pad_left, pad_right] |Attribute|int*|stride|[2]|[stride_y, stride_x] |Attribute|int*|dilation|[2]|[dilation_y, dilation_x] @@ -193,8 +193,8 @@ Performs a 3D convolution over the given input tensor. |Argument|Type|Name|Shape|Description |Input|in_t*|input|[N,ID,IH,IW,IC]|Input tensor -|Attribute|weight_t*|weight|[OC,KD,KH,KW,IC]|Weight kernel size KDxKHxKW -|Attribute|acc_t*|bias|[OC]|Per output channel bias data. +|Input (MT profile) Attribute (BI/MI profiles)|weight_t*|weight|[OC,KD,KH,KW,IC]|Weight kernel size KDxKHxKW +|Input (MT profile) Attribute (BI/MI profiles)|acc_t*|bias|[OC]|Per output channel bias data. |Attribute|int*|pad|[6]|[pad_d0, pad_d1, pad_top, pad_bottom, pad_left, pad_right] |Attribute|int*|stride|[3]|[stride_d, stride_y, stride_x] |Attribute|int*|dilation|[3]|[dilation_d, dilation_y, dilation_x] @@ -257,8 +257,8 @@ Performs 2D convolutions separately over each channel of the given tensor input, |Argument|Type|Name|Shape|Description |Input|in_t*|input|[N,H,W,C]|Input tensor -|Attribute|weight_t*|weight|[KH,KW,C,M]|Weight kernel size KH x KW -|Attribute|acc_t*|bias|[C*M]|Per output channel bias data. +|Input (MT profile) Attribute (BI/MI profiles)|weight_t*|weight|[KH,KW,C,M]|Weight kernel size KH x KW +|Input (MT profile) Attribute (BI/MI profiles)|acc_t*|bias|[C*M]|Per output channel bias data. |Attribute|int*|pad|[4]|[pad_top, pad_bottom, pad_left, pad_right] |Attribute|int*|stride|[2]|[stride_y, stride_x] |Attribute|int*|dilation|[2]|[dilation_y, dilation_x] @@ -467,8 +467,8 @@ Performs a 2D transposed convolution over the given tensor input, using the weig |Argument|Type|Name|Shape|Description |Input|in_t*|input|[N,IH,IW,IC]|Input tensor -|Attribute|weight_t*|weight|[OC,KH,KW,IC]|Weight kernel size KH x KW -|Attribute|acc_t*|bias|[OC]|Per output channel bias data. +|Input (MT profile) Attribute (BI/MI profiles)|weight_t*|weight|[OC,KH,KW,IC]|Weight kernel size KH x KW +|Input (MT profile) Attribute (BI/MI profiles)|acc_t*|bias|[OC]|Per output channel bias data. |Attribute|int*|outpad|[2]|[outpad_top, outpad_left] |Attribute|int*|stride|[2]|[stride_y, stride_x] |Attribute|int*|dilation|[2]|[dilation_y, dilation_x] diff --git a/chapters/type_conversion.adoc b/chapters/type_conversion.adoc index 6d3bcd8..2689818 100644 --- a/chapters/type_conversion.adoc +++ b/chapters/type_conversion.adoc @@ -91,9 +91,9 @@ Rescale quantized values into a new domain. This function scales by factor: mult |Attribute|in_t|input_zp|-|Input tensor zero point |Attribute|out_t|output_zp|-|Output tensor zero point -|Attribute|mul_t|multiplier[NC]|-|Scaling multiplier array -|Attribute|uint6_t|shift[NC] |-|Scaling shift array -|Attribute|bool|scale32|-|if (scale32) mul_t=int32_t else mul_t=int16_t +|Input (MT profile) Attribute (BI/MI profiles)|mul_t|multiplier[NC]|-|Scaling multiplier array +|Input (MT profile) Attribute (BI/MI profiles)|uint6_t|shift[NC] |-|Scaling shift array +|Input (MT profile) Attribute (BI/MI profiles)|bool|scale32|-|if (scale32) mul_t=int32_t else mul_t=int16_t |Attribute|bool|double_round|-|Select double round mode |Attribute|bool|per_channel|-|if (per_channel) NC=shape[dims-1] else NC=1 |=== |