Parameter fl meaning in .quantize file

What does the parameter fl mean in .quantize file produced after the shell conversion script is run. How can one calculate it? Why it is changing? Does it depend on dtype?

%YAML 1.2
---
# !!!This file disallow TABs!!!
quantize_parameters:
    '@input_0:out0':
        dtype: dynamic_fixed_point
        method: layer
        max_value:
        -   0.99609375
        min_value:
        -   0.0
        fl:
        -   7
        qtype: i8
    '@convolution_0_1:out0':
        dtype: dynamic_fixed_point
        method: layer
        max_value:
        -   21.542137145996094
        min_value:
        -   -13.79297161102295
        fl:
        -   2
        qtype: i8
    '@leakyrelu_0_3:out0':
        dtype: dynamic_fixed_point
        method: layer
        max_value:
        -   21.542137145996094
        min_value:
        -   -1.379297137260437
        fl:
        -   2
        qtype: i8
    '@convolution_1_4:out0':
        dtype: dynamic_fixed_point
        method: layer
        max_value:
        -   10.411376953125
        min_value:
        -   -15.914185523986816
        fl:
        -   3
        qtype: i8

@Frank

@enggsajjad

I don’t have much research on this file, you can change the parameters to make some comparisons.

The guide Model_Transcoding and Running User Guide_V0.8 explains this:

Int8: indicates that 32-bit floating point numbers are quantized to int8 data. Int8 data
includes eight bits, among which one is the sign bit and the other seven are valid
numbers. You need to calculate the fl value that indicates the number of bits in the
decimal. (The mechanism for int16 is similar to that of int8).