Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

Enhance gpu quantization #14094

Merged
merged 12 commits into from
Mar 6, 2019
Merged

Enhance gpu quantization #14094

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
17415db
enhance gpu quantization
Feb 8, 2019
a240ec9
fix test and improve error message
Feb 8, 2019
519da15
resolve conflict
Feb 19, 2019
6b06d50
add check srctype to quantized_conv.cu
Feb 25, 2019
00d8099
improve infer type
Feb 26, 2019
44d959e
fix lint
Mar 1, 2019
ab68668
add dtype check in quantize
Mar 5, 2019
8406726
revert check in python level and quantized_conv
Mar 5, 2019
7b43c18
Revert "add dtype check in quantize"
Mar 5, 2019
71999ef
Merge remote-tracking branch 'upstream/master' into enhance-gpu-quant…
Mar 5, 2019
845c063
add dtype check in quantize
Mar 5, 2019
9c44eb5
fix quantize test case
Mar 5, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions python/mxnet/contrib/quantization.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -499,6 +499,9 @@ def quantize_model(sym, arg_params, aux_params,
if quantized_dtype not in ('int8', 'uint8'):
raise ValueError('unknown quantized_dtype %s received,'
' expected `int8` or `uint8`' % quantized_dtype)
if quantized_dtype == 'uint8' and ctx != cpu():
raise ValueError('currently, uint8 quantization is only supported by CPU,'
' please switch to the context of CPU or int8 data type for GPU')
Copy link
Member

@anirudh2290 anirudh2290 Feb 9, 2019 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

better to add this error to backend like in the case for MKLDNN with int8 so that we dont have to add error handling to other frontends when we support quantization.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

currently, only python frontend support quantization and in fact calibration progress will not use backend specific quantized operator. So I think it's good to add error message in this place currently.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In QuantizeCompute (quantize-inl.h) you can check if std::is_same<xpu,gpu>::value and check for param.out_type and throw exception.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think this modification can work since infer type error mxnet.base.MXNetError: [02:07:55] /home/ubuntu/experimentals/1.4_release/src/operator/quantization/../tensor/matrix_op-inl.h:250: Check failed: src.type_flag_ == ret.type_flag_ (3 vs. 5) will occur before QuantizeCompute and we cannot get the ctx information during infer stage. So I think it's good to interrupt this action during the calibration stage.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

isnt that called from the forward pass of quantized_conv ? The quantize forward pass should execute before this.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

add check src_type in quantized_conv.cu, please take a review again.

qsym = _quantize_symbol(sym, excluded_symbols=excluded_sym_names,
offline_params=list(arg_params.keys()),
quantized_dtype=quantized_dtype,
Expand Down
10 changes: 10 additions & 0 deletions tests/python/quantization/test_quantization.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -450,6 +450,16 @@ def get_fp32_sym_with_multiple_outputs(length=1):
@with_seed()
def test_quantize_model():
def check_quantize_model(qdtype):
if is_test_for_native_cpu():
print('skipped testing quantize_model for native cpu since it is not supported yet')
return
elif qdtype == 'int8' and is_test_for_mkldnn():
print('skipped testing quantize_model for mkldnn cpu int8 since it is not supported yet')
return
elif qdtype == 'uint8' and is_test_for_gpu():
print('skipped testing quantize_model for gpu uint8 since it is not supported yet')
return
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please add else clause.


def check_params(params, qparams, qsym=None):
if qsym is None:
assert len(params) == len(qparams)
Expand Down