| /external/tensorflow/tensorflow/lite/toco/ |
| D | types.proto | 23 // Float32, not quantized
26 // Uint8, quantized
29 // Int32, not quantized
32 // Int64, not quantized
35 // String, not quantized
38 // Int16, quantized
44 // Complex64, not quantized
47 // Int8, quantized based on QuantizationParameters in schema.
|
| D | toco_flags.proto | 52 // by default input arrays are quantized like other arrays.
55 // we mean float arrays, and quantized arrays. This excludes plain
60 // inference. For such image models, the uint8 input is quantized, i.e.
71 // we mean float arrays, and quantized arrays. This excludes plain
81 // the output file. If they were quantized in the input file, then
83 // - If QUANTIZED_UINT8, then real-numbers arrays will be quantized
85 // then they get quantized.
96 // to estimate the performance of quantized inference, without caring about
104 // they make it easy to quantize models, but the resulting quantized model
107 // These values only apply to arrays quantized with the kUint8 data type.
[all …]
|
| /external/tensorflow/tensorflow/core/api_def/base_api/ |
| D | api_def_QuantizedMul.pbtxt | 6 The float value that the lowest quantized `x` value represents.
12 The float value that the highest quantized `x` value represents.
18 The float value that the lowest quantized `y` value represents.
24 The float value that the highest quantized `y` value represents.
30 The float value that the lowest quantized output value represents.
36 The float value that the highest quantized output value represents.
42 summary: "Returns x * y element-wise, working on quantized buffers."
|
| D | api_def_QuantizedAdd.pbtxt | 6 The float value that the lowest quantized `x` value represents.
12 The float value that the highest quantized `x` value represents.
18 The float value that the lowest quantized `y` value represents.
24 The float value that the highest quantized `y` value represents.
30 The float value that the lowest quantized output value represents.
36 The float value that the highest quantized output value represents.
42 summary: "Returns x + y element-wise, working on quantized buffers."
|
| D | api_def_QuantizedBatchNormWithGlobalNormalization.pbtxt | 12 The value represented by the lowest quantized input.
18 The value represented by the highest quantized input.
32 The value represented by the lowest quantized mean.
38 The value represented by the highest quantized mean.
52 The value represented by the lowest quantized variance.
58 The value represented by the highest quantized variance.
71 The value represented by the lowest quantized offset.
77 The value represented by the highest quantized offset.
91 The value represented by the lowest quantized gamma.
97 The value represented by the highest quantized gamma.
|
| D | api_def_QuantizedConv2D.pbtxt | 12 The float value that the lowest quantized input value represents.
18 The float value that the highest quantized input value represents.
24 The float value that the lowest quantized filter value represents.
30 The float value that the highest quantized filter value represents.
36 The float value that the lowest quantized output value represents.
42 The float value that the highest quantized output value represents.
68 summary: "Computes a 2D convolution given quantized 4D input and filter tensors."
70 The inputs are quantized tensors where the lowest value represents the real
72 This means that you can only interpret the quantized output in the same way, by
|
| D | api_def_Requantize.pbtxt | 6 The float value that the minimum quantized input value represents.
12 The float value that the maximum quantized input value represents.
18 The float value that the minimum quantized output value represents.
24 The float value that the maximum quantized output value represents.
52 "Converts the quantized `input` tensor into a lower-precision `output`."
54 Converts the quantized `input` tensor into a lower-precision `output`, using the
59 `input_max` is 1.0f, and we are dealing with `quint16` quantized data, then a 0
|
| D | api_def_QuantizedBiasAdd.pbtxt | 12 The float value that the lowest quantized input value represents.
18 The float value that the highest quantized input value represents.
24 The float value that the lowest quantized bias value represents.
30 The float value that the highest quantized bias value represents.
36 The float value that the lowest quantized output value represents.
42 The float value that the highest quantized output value represents.
|
| D | api_def_QuantizedMatMul.pbtxt | 18 The float value that the lowest quantized `a` value represents.
24 The float value that the highest quantized `a` value represents.
30 The float value that the lowest quantized `b` value represents.
36 The float value that the highest quantized `b` value represents.
42 The float value that the lowest quantized output value represents.
48 The float value that the highest quantized output value represents.
70 summary: "Perform a quantized matrix multiplication of `a` by the matrix `b`."
|
| D | api_def_QuantizeV2.pbtxt | 18 The quantized data produced from the float input.
37 used to convert the float values to their quantized equivalents. The
39 when rounding float values to their quantized equivalents.
68 quantized = round(input * range_scale) - round(range_min * range_scale) +
70 quantized = max(quantized, numeric_limits<T>::min())
71 quantized = min(quantized, numeric_limits<T>::max())
129 they will be separated by a small epsilon value to prevent ill-formed quantized
131 quantized values map to the same float value, which causes problems for
|
| D | api_def_QuantizedAvgPool.pbtxt | 12 The float value that the lowest quantized input value represents.
18 The float value that the highest quantized input value represents.
24 The float value that the lowest quantized output value represents.
30 The float value that the highest quantized output value represents.
53 summary: "Produces the average pool of the input tensor for quantized types."
|
| D | api_def_QuantizedMaxPool.pbtxt | 12 The float value that the lowest quantized input value represents.
18 The float value that the highest quantized input value represents.
24 The float value that the lowest quantized output value represents.
30 The float value that the highest quantized output value represents.
53 summary: "Produces the max pool of the input tensor for quantized types."
|
| D | api_def_QuantizedRelu.pbtxt | 6 The float value that the lowest quantized value represents.
12 The float value that the highest quantized value represents.
24 The float value that the lowest quantized value represents.
30 The float value that the highest quantized value represents.
|
| D | api_def_QuantizedRelu6.pbtxt | 6 The float value that the lowest quantized value represents.
12 The float value that the highest quantized value represents.
24 The float value that the lowest quantized value represents.
30 The float value that the highest quantized value represents.
|
| D | api_def_QuantizedReluX.pbtxt | 6 The float value that the lowest quantized value represents.
12 The float value that the highest quantized value represents.
24 The float value that the lowest quantized value represents.
30 The float value that the highest quantized value represents.
|
| D | api_def_RequantizationRange.pbtxt | 6 The float value that the minimum quantized input value represents.
12 The float value that the maximum quantized input value represents.
34 "Computes a range that covers the actual values present in a quantized tensor."
36 Given a quantized tensor described by `(input, input_min, input_max)`, outputs a
|
| D | api_def_QuantizeDownAndShrinkRange.pbtxt | 6 The float value that the minimum quantized input value represents.
12 The float value that the maximum quantized input value represents.
18 The float value that the minimum quantized output value represents.
24 The float value that the maximum quantized output value represents.
39 summary: "Convert the quantized \'input\' tensor into a lower-precision \'output\', using the"
46 input_max is 1.0f, and we are dealing with quint16 quantized data, then a 0
|
| D | api_def_QuantizedInstanceNorm.pbtxt | 12 The value represented by the lowest quantized input.
18 The value represented by the highest quantized input.
30 The value represented by the lowest quantized output.
36 The value represented by the highest quantized output.
|
| D | api_def_QuantizedResizeBilinear.pbtxt | 30 summary: "Resize quantized `images` to `size` using quantized bilinear interpolation."
32 Input images and output images must be quantized types.
|
| D | api_def_QuantizedConcat.pbtxt | 40 The float value that the minimum quantized output value represents.
46 The float value that the maximum quantized output value represents.
49 summary: "Concatenates quantized tensors along one dimension."
|
| D | api_def_QuantizeAndDequantizeV2.pbtxt | 48 used when rounding float values to their quantized equivalents. The following
59 This op simulates the precision loss from the quantized forward pass by:
69 From the specified 'num_bits' in the quantized output type, it determines
70 minimum and maximum representable quantized values.
88 quantized range.
|
| /external/tensorflow/tensorflow/lite/g3doc/convert/ |
| D | cmdline_reference.md | 70 is quantized. This is only needed if `inference_input_type` is
73 quantized value in the quantized input array will be interpreted as a
78 quantized input, the quantized input would be immediately dequantized by
81 * When performing quantized inference
86 quantized inference code. `mean_value` must be an integer when
87 performing quantized inference.
99 This flag only impacts real-number arrays including float and quantized
104 file. If they were quantized in the input file, then they get
106 * If `QUANTIZED_UINT8`, then real-numbers arrays will be quantized as
108 they get quantized.
[all …]
|
| /external/tensorflow/tensorflow/java/src/gen/resources/ |
| D | tftypes.csv | 12 QInt8,n,,n,,quantized int8
13 QUInt8,n,,n,,quantized uint8
14 QInt32,n,,n,,quantized int32
16 QInt16,n,,n,,quantized int16
17 QUInt16,n,,n,,quantized uint16
|
| /external/gemmlowp/doc/ |
| D | quantization.md | 11 of quantized 8bit values and matrices of real numbers. The choice of a
20 output pipeline, allowing the user to implement their own preferred quantized
36 quantized uint8 values, the mapping between real and quantized uint8 values must
50 mean that the result of the quantized calculations do no longer readily provide
64 In order for such zero-padding to be feasible in a quantized implementation of
66 quantized form, i.e. that it correspond exactly to some quantized value, which
70 quantized value for padding, that does not exactly correspond to the real value
130 * `zero_point` is integral, more specifically is one of the possible quantized
139 to quantized values (typically uint8), we turn to applying this knowledge to
141 multiplication of matrices of quantized values.
[all …]
|
| /external/tensorflow/tensorflow/lite/g3doc/performance/ |
| D | post_training_quantization.md | 32 Since weights are quantized post training, there could be an accuracy loss, particularly for
33 smaller networks. Pre-trained fully quantized models are provided for specific networks in
34 …l repository](../models/). It is important to check the accuracy of the quantized model to verify …
39 ### Representation for quantized tensors
54 values corresponding to the lowest and highest quantized value. Each entry in the
55 quantized array represents a float value in that range, distributed linearly
64 8-bit array, the quantized values would be represented as follows:
|