quantization aware training pass #3817
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…ing_ops' into quant_aware_training_dqx
Ldpe2G
reviewed
Nov 13, 2020
Ldpe2G
reviewed
Nov 15, 2020
Ldpe2G
reviewed
Nov 16, 2020
|
|
||
| namespace { | ||
|
|
||
| void VerifyQATList(const QATList& amp_list) { |
There was a problem hiding this comment.
void VerifyQATList(const QATList& qat_list) {
Ldpe2G
reviewed
Nov 16, 2020
Ldpe2G
reviewed
Nov 16, 2020
Signed-off-by: daquexian <daquexian566@gmail.com>
Signed-off-by: daquexian <daquexian566@gmail.com>
Signed-off-by: daquexian <daquexian566@gmail.com>
daquexian
changed the title
Ldpe2G
approved these changes
Jan 13, 2021
daquexian
force-pushed
the
quant_aware_training_dqx
branch
from
047c9d6 to
3832433
Compare
daquexian
requested review from
oneflow-ci-bot
and removed request for
oneflow-ci-bot
oneflow-ci-bot
requested review from
oneflow-ci-bot
and removed request for
oneflow-ci-bot
Signed-off-by: daquexian <daquexian566@gmail.com>
liujuncheng
pushed a commit
that referenced
this pull request
Jun 3, 2021
* init qat pass * fix bugs * add calculate weight scale and zero_point op & unit tests * clear batch axis of scale and zero_point * add calculate activation scale and zero_point op & unit tests * add fake quantization ops & unit tests * add sbp signature to fake quantization op & improve code style * imporve unit test speed * update pass * add QatConfig * format * code clean * add calculate weight scale and zero_point op & unit tests * clear batch axis of scale and zero_point * add calculate activation scale and zero_point op & unit tests * add fake quantization ops & unit tests * add sbp signature to fake quantization op & improve code style * imporve unit test speed * make changes according to review comments * rename quantize ops following the pytorch's naming scheme * change the input zero_point of fake_quantize op to optional * stop updating moving_min and moving_max after training iteration reaches the given point * add calculate weight scale and zero_point op & unit tests * clear batch axis of scale and zero_point * add calculate activation scale and zero_point op & unit tests * add fake quantization ops & unit tests * add sbp signature to fake quantization op & improve code style * imporve unit test speed * make changes according to review comments * rename quantize ops following the pytorch's naming scheme * change the input zero_point of fake_quantize op to optional * stop updating moving_min and moving_max after training iteration reaches the given point * align with latest fake quant ops Signed-off-by: daquexian <daquexian566@gmail.com> * optimize CHECK Signed-off-by: daquexian <daquexian566@gmail.com> * add multiple devices tests && fix sbp infer error * fix bugs on mobilenetv2 Signed-off-by: daquexian <daquexian566@gmail.com> * align with latest fake quant ops Signed-off-by: daquexian <daquexian566@gmail.com> * format Signed-off-by: daquexian <daquexian566@gmail.com> * add calculate weight scale and zero_point op & unit tests * clear batch axis of scale and zero_point * add calculate activation scale and zero_point op & unit tests * add fake quantization ops & unit tests * add sbp signature to fake quantization op & improve code style * imporve unit test speed * make changes according to review comments * rename quantize ops following the pytorch's naming scheme * change the input zero_point of fake_quantize op to optional * stop updating moving_min and moving_max after training iteration reaches the given point * add multiple devices tests && fix sbp infer error * stop udpating moving max and min during the prediction mode * imporve ReduceMaxMinPerChannel cuda kernel slightly * align with cfg job_conf Signed-off-by: daquexian <daquexian566@gmail.com> * amp_lsit -> op_list Signed-off-by: daquexian <daquexian566@gmail.com> * support conv op with bias input Signed-off-by: daquexian <daquexian566@gmail.com> * add calculate weight scale and zero_point op & unit tests * clear batch axis of scale and zero_point * add calculate activation scale and zero_point op & unit tests * add fake quantization ops & unit tests * add sbp signature to fake quantization op & improve code style * imporve unit test speed * make changes according to review comments * rename quantize ops following the pytorch's naming scheme * change the input zero_point of fake_quantize op to optional * stop updating moving_min and moving_max after training iteration reaches the given point * add multiple devices tests && fix sbp infer error * stop udpating moving max and min during the prediction mode * imporve ReduceMaxMinPerChannel cuda kernel slightly * change quantize_to_bit to quantization_bit * change quantize to quantization * format Signed-off-by: daquexian <daquexian566@gmail.com> * fix bias zero point shape, add tests Signed-off-by: daquexian <daquexian566@gmail.com> * set 'training' attr according to job desc Signed-off-by: daquexian <daquexian566@gmail.com> * refine tests Signed-off-by: daquexian <daquexian566@gmail.com> * polish Signed-off-by: daquexian <daquexian566@gmail.com> * reformat Signed-off-by: daquexian <daquexian566@gmail.com> * fix cpu test Signed-off-by: daquexian <daquexian566@gmail.com> Co-authored-by: Ldpe2G <liangdepeng@gmail.com> Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com> Former-commit-id: a48c6e4
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
包含 quantization aware training 自动插入相关 op 的 pass