-
Notifications
You must be signed in to change notification settings - Fork 6.8k
Add quantization support for GluonCV #15754
Conversation
|
||
with TemporaryDirectory() as tmpdirname: | ||
prefix = os.path.join(tmpdirname, 'tmp') | ||
network.export(prefix, epoch=0) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I feel like it's more convenient to allow export return (sym, arg_params, aux_params) so we can avoid using temporary file.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It's no mandatory and urgent now, but gluoncv also has the same implementation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
agree, will enhance them later to make it more easier to switch between symbolic and block:)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
* enhance quantization api * integrate gluoncv solution * support gluon ssd * enhance api * [TODO]split to another PR * enhance example script * add wildcard match for exclude layers * support int8 dtype parameter * enable dataiter api * use try method * add unit test for quantize gluon * fix lint * fix lint 2 * fix temporary directory in python2 * fix lint * fix try import and add todo * trigger
Description
Add a User-level API for Gluon users to generate a quantized SymbolBlock from a FP32 HybridBlock w/ or w/o calibration.
@zhreshold @pengzhao-intel @TaoLv @ZhennanQin
Checklist
Essentials
Please feel free to remove inapplicable items for your PR.
Changes
Comments