Summary:
Continuing the work from https://github.com/pytorch/pytorch/pull/146427
Adds the `torch.float8_e8m0fnu` dtype to PyTorch, as detailed in
https://github.com/pytorch/pytorch/issues/146414 . Please see the issue for a detailed definition of the format. Example of basic functionality:
```python
import torch
# round trip
x0 = torch.randn(4, 4, dtype=torch.float32)
x1 = x0.to(torch.float8_e8m0fnu) # RNE rounding
x2 = x1.to(torch.float32) # 2 ** exponent
# creation with empty
x0 = torch.empty(4, 4, dtype=torch.float8_e8m0fnu)
# printing
print(x0)
```
Done in this PR:
* numerical correctness
* op coverage (except for `torch._scaled_mm`): create tensor, cast to/from float32
* printing a tensor works
For future PRs:
* performance optimizations for casting
* torch._scaled_mm
* PT2
* various cleanups (detailed in comments with issue numbers)
Test Plan:
```
pytest test/quantization/core/experimental/test_float8.py -s
```
Reviewers:
Subscribers:
Tasks:
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147466
Approved by: https://github.com/drisspg
Changes by apply order:
1. Replace all `".."` and `os.pardir` usage with `os.path.dirname(...)`.
2. Replace nested `os.path.dirname(os.path.dirname(...))` call with `str(Path(...).parent.parent)`.
3. Reorder `.absolute()` ~/ `.resolve()`~ and `.parent`: always resolve the path first.
`.parent{...}.absolute()` -> `.absolute().parent{...}`
4. Replace chained `.parent x N` with `.parents[${N - 1}]`: the code is easier to read (see 5.)
`.parent.parent.parent.parent` -> `.parents[3]`
5. ~Replace `.parents[${N - 1}]` with `.parents[${N} - 1]`: the code is easier to read and does not introduce any runtime overhead.~
~`.parents[3]` -> `.parents[4 - 1]`~
6. ~Replace `.parents[2 - 1]` with `.parent.parent`: because the code is shorter and easier to read.~
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129374
Approved by: https://github.com/justinchuby, https://github.com/malfet
Changes by apply order:
1. Replace all `".."` and `os.pardir` usage with `os.path.dirname(...)`.
2. Replace nested `os.path.dirname(os.path.dirname(...))` call with `str(Path(...).parent.parent)`.
3. Reorder `.absolute()` ~/ `.resolve()`~ and `.parent`: always resolve the path first.
`.parent{...}.absolute()` -> `.absolute().parent{...}`
4. Replace chained `.parent x N` with `.parents[${N - 1}]`: the code is easier to read (see 5.)
`.parent.parent.parent.parent` -> `.parents[3]`
5. ~Replace `.parents[${N - 1}]` with `.parents[${N} - 1]`: the code is easier to read and does not introduce any runtime overhead.~
~`.parents[3]` -> `.parents[4 - 1]`~
6. ~Replace `.parents[2 - 1]` with `.parent.parent`: because the code is shorter and easier to read.~
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129374
Approved by: https://github.com/justinchuby, https://github.com/malfet
Add option to split Linear gates for Quantizable LSTM into separate ops (#141366)
Summary:
Reattempt to land D65283170, adding pyre-fixmes / mypy ignores following D52890934
For LSTM, the input and hidden state are projected with Linear layers to construct the 4 gates. This is typically performed together as a single Linear (for each state) with output channel count `4 * hidden_dim` for efficiency.
https://www.internalfb.com/code/fbsource/[ebef7c4238aa55948b2b444044f2c8ed2040de55]/fbcode/caffe2/torch/ao/nn/quantizable/modules/rnn.py?lines=52-58
The output is then ultimately split into 4:
https://www.internalfb.com/code/fbsource/[ebef7c4238aa55948b2b444044f2c8ed2040de55]/fbcode/caffe2/torch/ao/nn/quantizable/modules/rnn.py?lines=83-87
For on-device latency (and possibly memory) considerations, we want to avoid constructing the intermediate `gates` tensor (which can be relatively large), by splitting `igates` and `hgates` first (as 4x `Linear(hidden_dim, hidden_dim)` each), applying add separately, then proceeding as usual.
This functionality can be enabled by specifying `split_gates=True` (default False is original behavior) at any entry point (directly with `torch.ao.nn.quantizable.LSTM` or via `_get_lstm_with_individually_observed_parts`).
Test Plan:
piggy back on existing test to check for correct swap handling, numerics, and jit.script during prepare/convert
```
buck2 test 'fbcode//mode/opt' fbcode//caffe2/test/quantization:test_quantization -- --exact 'caffe2/test/quantization:test_quantization - test_custom_module_lstm (caffe2.test.quantization.core.test_quantized_op.TestQuantizedOps)'
```
https://www.internalfb.com/intern/testinfra/testrun/4503599884152725
This test is quite long running now (more than double original).
---
shorter test to confirm original `LSTMCell` passes
```
buck2 test 'fbcode//mode/opt' fbcode//caffe2/test:quantization_fx -- --exact 'caffe2/test:quantization_fx - test_static_lstm_with_custom_fixed_qparams (quantization.fx.test_quantize_fx.TestQuantizeFx)'
```
https://www.internalfb.com/intern/testinfra/testrun/11258999127933996
Reviewed By: Ninja91
Differential Revision: D66380336
**About this PR**
This PR adds the following ops for `linear_dynamic_fp16` in onednn namespace. These ops are intended for PT2E quantization eager mode.
- `onednn::linear_prepack_fp16`: packs fp32 weight to an fp16 MkldnnCPU tensor.
- `onednn::linear_dynamic_fp16`: takes an fp32 CPU tensor and an fp16 MkldnnCPU tensor and compute linear in fp32
- `onednn::linear_relu_dynamic_fp16`: similar as the former and apply relu on output.
**Test plan**
`python test/test_quantization.py -k test_linear_dynamic_fp16_onednn`
**Implementation**
These ops call oneDNN lib under the hood. It's worth noting that oneDNN does not support f32 * f16 -> f32 computation, so we have to convert fp16 weight to fp32 before computation. And weight is still in plain format after packing.
**Correctness and performance**
Correctness is guaranteed by UT.
Performance of the new ops may be better than the FBGEMM implementation when weight shape is small but worse when weight shape is large. It's because weight dtype conversion and computation are not fused.
For example, I ran benchmarks on an Intel(R) Xeon(R) Platinum 8490H machine with different cores and shapes. When using 1 core per instance, the new implementation generally is faster for weight shape < 1024 * 1024. When using more cores, the threshold will increase.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/140376
Approved by: https://github.com/jerryzh168, https://github.com/jgong5
This PR adds C shim for `QConvPointWisePT2E` and `QConvPointWiseBinaryPT2E` similar to https://github.com/pytorch/pytorch/pull/138439. Besides that, we aligned the implementation of `qconv_pointwise` with `qlinear_pointwise` in the following aspects:
1. The parameter order of `qconv_pointwise` and `qlinear_pointwise` are quite different, we aligned the schema of `qconv_pointwise` to have similar parameter order as `qlinear_pointwise` to make it more consistent.
2. We always converted `x_scale` and `x_zero_point` to Tensors, just like in the lowering of `qlinear_pointwise`. This avoids the need to create two separate C APIs (one for `double x_scale` and `int64_t x_zero_point`, and another for `Tensor` versions). Instead, we only need one API for `Tensor`-based `x_scale` and `x_zero_point`. If we later add dynamic quantization for qconv (which will use `Tensor` for `x_scale` and `x_zero_point`), we can reuse the code from this PR and don't need to change the C shim layer API.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/138540
Approved by: https://github.com/jgong5, https://github.com/desertfire
ghstack dependencies: #138691, #138806
Summary:
This diff adds two new operators torch.ops._quantized.wrapped_linear_prepack and torch.ops._quantized.wrapped_quantized_linear_prepacked. It is a decomposition of the op torch.ops._quantized.wrapped_quantized_linear added in the previous diff.
We decomposed in this way as packed weight could be computed early so we don;t need to do it in every forward in AOTI
Reviewed By: jerryzh168
Differential Revision: D61395887
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134232
Approved by: https://github.com/houseroad
Summary:
This diff adds a new operator wrapped_quantized_linear (torch.ops._quantized.wrapped_quantized_linear) and takes the following input argument: input (in fp32) , input_scale, input_zero_point, weight (in fp32), weight_scale, weight_zero_point, bias (in fp32), output_scale, output_zero_point, and out_channel. It does the following
1. Use quantize_per_tensor(input, input_scale, input_zero_point) to quantize the input tensor to int8
2. Use quantized::linear_prepack(weight, weight_scale, weight_zero_point, bias) to pack the weight and bias
3. Use quantized::linear to perform int8 quantized linear
4. dequantize
This new op is essentially a wrapper of mutiple ops. We do this as torch.export cannot handle models where it has old quantize apis.
Reviewed By: jerryzh168
Differential Revision: D61377266
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134024
Approved by: https://github.com/houseroad
Summary:
This diff aims to fix the GPU Test skips in the quantization tests under the `caffe2/test/quantization` directory. The changes made in the `TARGETS` files include adding the `should_use_remote_gpu` flag to enable remote GPU testing. This should help to resolve the skipped tests and improve the overall test coverage.
[This diff] Fixed skip count: 4
[Running total] Fixed skip count: 4
Note: Creating separate diffs for each test-group.
Test Plan:
**281475054644766**: buck2 test 'fbcode//mode/opt' fbcode//caffe2/test/quantization:test_quantization -- --exact 'caffe2/test/quantization:test_quantization - test_compare_per_channel_device_numerics (caffe2.test.quantization.core.test_quantized_tensor.TestQuantizedTensor)'
https://www.internalfb.com/intern/testinfra/testrun/5629499773981783
**281475054644780**: buck2 test 'fbcode//mode/opt' fbcode//caffe2/test/quantization:test_quantization -- --exact 'caffe2/test/quantization:test_quantization - test_compare_per_tensor_device_numerics (caffe2.test.quantization.core.test_quantized_tensor.TestQuantizedTensor)'
https://www.internalfb.com/intern/testinfra/testrun/11540474087422107
**281475054644853**: buck2 test 'fbcode//mode/opt' fbcode//caffe2/test/quantization:test_quantization -- --exact 'caffe2/test/quantization:test_quantization - test_quant_pin_memory (caffe2.test.quantization.core.test_quantized_tensor.TestQuantizedTensor)'
https://www.internalfb.com/intern/testinfra/testrun/11540474087422477
**844425008078016**: buck2 test 'fbcode//mode/opt' fbcode//caffe2/test/quantization:test_quantization -- --exact 'caffe2/test/quantization:test_quantization - test_cuda_quantization_does_not_pin_memory (caffe2.test.quantization.core.test_quantized_tensor.TestQuantizedTensor)'
https://www.internalfb.com/intern/testinfra/testrun/1407375259845199
Differential Revision: D60055277
Pull Request resolved: https://github.com/pytorch/pytorch/pull/133158
Approved by: https://github.com/jovianjaison
Add similar semantics for creating a buffer object similar to creating a parameter. This is done by introducing a new Buffer class that can be used for type disambiguation. The underlying functionality of registering a buffer remains the same as the register_buffer method has not been changed. The persistent parameter in the Buffer type is to indicate whether a buffer object should be persistent or not. Other non-test changes have to do with getting the new Buffer type recognized by inductor and dynamo. Remaining changes are test changes to make sure that the Buffer type can be used as a drop in replacement for register_buffer as it just leads to register_buffer being called. The addition of this new functionality still allows for normal tensors to be used as buffers so these changes are intended to be backwards compatible.
Fixes#35735
Co-authored-by: Mikayla Gawarecki <mikaylagawarecki@gmail.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/125971
Approved by: https://github.com/albanD, https://github.com/anijain2305, https://github.com/mlazos
Summary:
There were two problems with the HistogramObserver:
1. It does not work when someone passes a batch_size 1, tensor_size 1 data-point.
2. The Histogram doesn't seem to actually update if the range of the new x falls within the old one
These issues were both fixed.
On top of this, I greatly simplified the logic for the histogram updating. Now, it doesn't do the downsampling anymore, which saves a ton of memory and code. The accuracy can still be controlled with the upsampling ratio. This ratio was also too high for the accuracy we generally need here, I reduced the default for this.
Also the code is cleaner now, much easier to follow what's happening.
test_histogram_observer_same_inputs was likely wrong - If I pass 0s and 1s to my histogramobserver, I want them to actually count! The current test now thinks it's good to discard and ignore these values.
Test Plan: You can run the included tests.
Differential Revision: D58931336
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129387
Approved by: https://github.com/jerryzh168
**Summary**
We change the schema of QLinear Binary, so it will be easier to enable the corresponding gemm template.
- Extra input of binary post-op is a tensor which needs to be an input node of autotuning, we need to move it at front of `output_scale` which is a scalar.
- We also move it at front of `bias`, since `bias` is optional tensor for this fusion, but `other` is a must to have for linear binary fusion.
**Test Plan**
```
python -u -m pytest -s -v test/quantization/core/test_quantized_op.py -k qlinear
python -u -m pytest -s -v test/inductor/test_mkldnn_pattern_matcher.py -k qlinear
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129049
Approved by: https://github.com/jgong5, https://github.com/jansel
ghstack dependencies: #128825, #129048
Changes by apply order:
1. Replace all `".."` and `os.pardir` usage with `os.path.dirname(...)`.
2. Replace nested `os.path.dirname(os.path.dirname(...))` call with `str(Path(...).parent.parent)`.
3. Reorder `.absolute()` ~/ `.resolve()`~ and `.parent`: always resolve the path first.
`.parent{...}.absolute()` -> `.absolute().parent{...}`
4. Replace chained `.parent x N` with `.parents[${N - 1}]`: the code is easier to read (see 5.)
`.parent.parent.parent.parent` -> `.parents[3]`
5. ~Replace `.parents[${N - 1}]` with `.parents[${N} - 1]`: the code is easier to read and does not introduce any runtime overhead.~
~`.parents[3]` -> `.parents[4 - 1]`~
6. ~Replace `.parents[2 - 1]` with `.parent.parent`: because the code is shorter and easier to read.~
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129374
Approved by: https://github.com/justinchuby, https://github.com/malfet
Changes by apply order:
1. Replace all `".."` and `os.pardir` usage with `os.path.dirname(...)`.
2. Replace nested `os.path.dirname(os.path.dirname(...))` call with `str(Path(...).parent.parent)`.
3. Reorder `.absolute()` ~/ `.resolve()`~ and `.parent`: always resolve the path first.
`.parent{...}.absolute()` -> `.absolute().parent{...}`
4. Replace chained `.parent x N` with `.parents[${N - 1}]`: the code is easier to read (see 5.)
`.parent.parent.parent.parent` -> `.parents[3]`
5. ~Replace `.parents[${N - 1}]` with `.parents[${N} - 1]`: the code is easier to read and does not introduce any runtime overhead.~
~`.parents[3]` -> `.parents[4 - 1]`~
6. ~Replace `.parents[2 - 1]` with `.parent.parent`: because the code is shorter and easier to read.~
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129374
Approved by: https://github.com/justinchuby, https://github.com/malfet
