for #1606
draft #1924
Note: I wasn't able to create this operator via tflite converter.
For some reason, TF2.2 converts tf.math.round as a custom operator.
As I wrote in https://github.com/Samsung/ONE/issues/1606#issuecomment-642935352, I'd like to continue this work cause of schedule issue
node {
name: "round"
op: "Round"
input: "Placeholder"
attr {
key: "T"
value {
type: DT_FLOAT
}
}
}
Operator Codes: [order] OpCodeName (OpCode Enum)
[0] ROUND (code: 116, version: 1)
Buffers: B(index) (length) values, if any
B(0) (0)
B(1) (0)
B(2) (0)
B(3) (16) 0x31 0x2e 0x31 0x34 0x2e 0x30 00 00 00 00 00 00 00 00 00 00 ...
-------------------------------------------------------------------
Sub-Graph: #0 main
Operands: T(subgraph index : tensor index) TYPE (shape) (shape_signature) B(buffer index) OperandName
T(0:0) FLOAT32 (2, 3) B(1) Placeholder
T(0:1) FLOAT32 (2, 3) B(2) round
Operators: O(subgraph index : operator index) OpCodeName
Option(values) ... <-- depending on OpCode
I T(tensor index) OperandName <-- as input
O T(tensor index) OperandName <-- as output
O(0:0) ROUND
I T(0:0) Placeholder
O T(0:1) round
Inputs/Outputs: I(input)/O(output) T(tensor index) OperandName
I T(0:0) Placeholder
O T(0:1) round
For some reason, TF2.2 converts tf.math.round as a custom operator.
I think you wasn't running in TF 2.2. it may been 1.x
I think you wasn't running in TF 2.2. it may been 1.x
Indeed. I checked, It was 2.0.0.b1 =_=
@seanshpark FYI
I rebased recipe in #2004
You can merge it if you want (or not, if it is faster to make it yourself)
Everything is done, closing
Thank you!
Most helpful comment
Everything is done, closing