Open hseok-oh opened 3 years ago
@hseok-oh As you already know, some operators need to be implemented in priority because they are used in one of our target model.
Note that Conv2D
and DepthwiseConv2D
use channel-wise quantization.
Also we need to take a look
because it got segmentation fault during running QUANTIZE
only nnpackage.
$ LD_LIBRARY_PATH=lib Product/out/bin/nnpackage_run --nnpackage quantize
Package Filename quantize
Segmentation fault
(ADDED)
I will take a look QUANTIZE
, then get started with MUL
and SUB
.
(ADDED)
I've found there is problem selecting quantize
from our target model by select_operator.py
.
I've tried to fix it, however, it is another problem. I will make another issue with current investigation.
I've continued with workaround (#5353). Now, it shows:
$ LD_LIBRARY_PATH=lib Product/out/bin/nnpackage_run --nnpackage quantize
Package Filename quantize
Error during model loading : OperationValidator failed at line 288
Quantize
does not handle INT8
input.
I will start to enable INT8
input type for Quantize
since Quantize
is the first operator of our target model.
I've created a PR #5446 for Quantize
. Today, I started MUL
— the next operator. I will do SUB
and ADD
together.
I'll start SOFTMAX
and RESIZE_BILINEAR
.
I'll start Conv2D
. It requires handling of channel-wise quantization params. (loader, IR, ...).
I'll start PAD
.
@glistening I'll start AVERAGE_POOL_2D
Prerequest
NYI cpu backend kernel: int8 quantization type