Summary:
This diff adds FakeQuantizeWithBackward. This works the same way as the regular FakeQuantize module, allowing QAT to occur in the forward pass, except it has an additional quantize_backward parameter. When quantize_backward is enabled, the gradients are fake quantized as well (dynamically, using hard-coded values). This allows the user to see whether there would be a significant loss of accuracy if the gradients were quantized in their model.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40532
Test Plan: The relevant test for this can be run using `python test/test_quantization.py TestQATBackward.test_forward_and_backward`
Reviewed By: supriyar
Differential Revision: D22217029
Pulled By: durumu
fbshipit-source-id: 7055a2cdafcf022f1ea11c3442721ae146d2b3f2
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/42348
Use the dtype info in placeholderObserver to decide what ops to insert in the graph
In the next PR we can delete NoopObserver
Test Plan:
python test/test_quantization.py
Imported from OSS
Reviewed By: jerryzh168
Differential Revision: D22859457
fbshipit-source-id: a5c618f22315534ebd9a2df77b14a0aece196989
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/42222
This change adds the necessary passes to perform FP16 dynamic quantization.
We skip inserting observers for activations based on the dtype (torch.float16) and only insert the Fp16Observer for weights
Test Plan:
python test/test_quantization.py TestQuantizeJitOps
Imported from OSS
Reviewed By: jerryzh168
Differential Revision: D22849220
fbshipit-source-id: 2c53594ecd2485e9e3dd0b380eceaf7c5ab5fc50
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39874
When fbgemm backend is set we make sure reduce_range is set to true to avoid overflow in the operator
Also adds test for per-channel quant with graph mode and compare numerics with eager mode
Test Plan:
python test/test_quantization.py
Imported from OSS
Differential Revision: D22011205
fbshipit-source-id: 1c7c9b7ab0d84200e3d8d85c34978554c30c0169
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38407
We can still run some quantized tests even when fbgemm/qnnpack isn't enabled
Test Plan:
python test/test_quantization.py
Imported from OSS
Differential Revision: D21554257
fbshipit-source-id: e4fa8f61f6a6717881c00620ed7938c01ffbf958
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35455
In graph mode we need to observer the activation tensor for dynamic quantization. This observer should behave the same way as the quantization functions called in the dynamic operator.
Currently for qlinear_dynamic we call quant_utils::ChooseQuantizationParams which has its own logic for calculating scale and zero_point.
We mimic those calculations in the new observer.
Test Plan:
python test/test_quantization.py ObserverTest
Imported from OSS
Differential Revision: D20664586
fbshipit-source-id: e987ea71fff777c21e00c498504e6586e92568a2
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35265
In graph mode we need to observer the activation tensor for dynamic quantization. This observer should behave the same way as the quantization functions called in the dynamic operator.
Currently for qlinear_dynamic we call quant_utils::ChooseQuantizationParams which has its own logic for calculating scale and zero_point.
We mimic those calculations in the new observer.
Test Plan:
python test/test_quantization.py ObserverTest
Imported from OSS
Differential Revision: D20630988
fbshipit-source-id: 7e7aca77590f965dcb423a705e68d030aaf98550
Summary:
The PR tried to enable the per-channel(row-wise) dynamic quantization for linear operator. Given we have seen some accuracy drop due to the per-tensor quantization, we expect the per-channel could help improve the accuracy.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30122
Differential Revision: D18630541
Pulled By: lly-zero-one
fbshipit-source-id: d52685deec5e7de46cd686ae649a8c8765b9cacf
