Refactor (linter): add support for isort (#505)

* chore: pre-commit is now a requirement

* chore: support for isort for import sorting

* chore: isort linted files

* CI: pre-commit checks in github actions

* Pre-commit fix, workflow works

.github/workflows/develop_install.yml

@@ -52,13 +52,11 @@ jobs:
       if: startsWith(runner.os, 'macOS') == true
 
     - name: Run Nox session for testing brevitas develop install and imports
-      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}} 
+      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}}
       shell: bash
       run: nox -v -s tests_brevitas_install_dev-${{ matrix.python_version }}\(\pytorch_${{ matrix.pytorch_version }}\)
 
     - name: Run Nox session for testing brevitas_examples develop install and imports
-      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}} 
+      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}}
       shell: bash
       run: nox -v -s tests_brevitas_examples_install_dev-${{ matrix.python_version }}\(\pytorch_${{ matrix.pytorch_version }}\)
-
-

.github/workflows/end_to_end.yml

@@ -56,8 +56,6 @@ jobs:
       if: startsWith(runner.os, 'macOS') == true
 
     - name: Run Nox session for end-to-end flows
-      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}} 
+      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}}
       shell: bash
       run: nox -v -s tests_brevitas_end_to_end-${{ matrix.python_version }}\(\pytorch_${{ matrix.pytorch_version }}\)
-
-

.github/workflows/examples_pytest.yml

@@ -53,8 +53,6 @@ jobs:
       if: startsWith(runner.os, 'macOS') == true
 
     - name: Run Nox session for brevitas_examples pytest
-      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}} 
+      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}}
       shell: bash
       run: nox -v -s tests_brevitas_examples_cpu-${{ matrix.python_version }}\(${{ matrix.jit_status }}\,\ pytorch_${{ matrix.pytorch_version }}\)
-
-

.github/workflows/finn_integration.yml

@@ -57,8 +57,6 @@ jobs:
       run: sudo apt-get install protobuf-compiler libprotoc-dev
 
     - name: Run Nox session for Brevitas-FINN integration
-      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}} 
+      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}}
       shell: bash
       run: nox -v -s tests_brevitas_finn_integration-${{ matrix.python_version }}\(\pytorch_${{ matrix.pytorch_version }}\)
-
-

.github/workflows/gen_github_actions.py

@@ -1,7 +1,7 @@
 from collections import OrderedDict as od
 
-from utils import Action, combine_od_list
-
+from utils import Action
+from utils import combine_od_list
 
 BASE_YML_TEMPLATE = 'base.yml.template'
 PYTEST_YML = 'pytest.yml'
@@ -208,4 +208,3 @@ def gen_test_brevitas_end_to_end():
     gen_test_brevitas_ort_integration()
     gen_test_brevitas_notebook()
     gen_test_brevitas_end_to_end()
-

.github/workflows/notebook.yml

@@ -59,8 +59,6 @@ jobs:
       if: startsWith(runner.os, 'macOS') == true
 
     - name: Run Nox session for Notebook execution
-      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}} 
+      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}}
       shell: bash
       run: nox -v -s tests_brevitas_notebook-${{ matrix.python_version }}\(\pytorch_${{ matrix.pytorch_version }}\)
-
-

.github/workflows/ort_integration.yml

@@ -52,8 +52,6 @@ jobs:
       if: startsWith(runner.os, 'macOS') == true
 
     - name: Run Nox session for Brevitas-ORT integration
-      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}} 
+      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}}
       shell: bash
       run: nox -v -s tests_brevitas_ort_integration-${{ matrix.python_version }}\(\pytorch_${{ matrix.pytorch_version }}\)
-
-

.github/workflows/pre_commit.yml

@@ -0,0 +1,32 @@
+name: Pre-Commit Workflow
+
+on:
+  push:
+    branches: [ master, dev ]
+  pull_request:
+    types:
+      - opened
+      - reopened
+      - synchronize
+      - ready_for_review
+      - review_requested
+
+jobs:
+  pre_commit:
+    name: pre-commit
+    runs-on: ubuntu-latest
+
+    if: ${{ !github.event.pull_request.draft }}
+    steps:
+    - name: Checkout
+      uses: actions/checkout@v3
+
+    - name: Install Python
+      uses: actions/setup-python@v4
+      with:
+        python-version: '3.7'
+
+    - name: Run pre-commit
+      uses: pre-commit/[email protected]
+      with:
+        extra_args: --verbose --all-files

.github/workflows/pytest.yml

@@ -53,8 +53,6 @@ jobs:
       if: startsWith(runner.os, 'macOS') == true
 
     - name: Run Nox session for brevitas pytest
-      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}} 
+      if: ${{ (contains(fromJson('["1.5.1", "1.10.1", "1.13.0"]'), matrix.pytorch_version ) && github.event_name == 'pull_request' && matrix.platform == 'ubuntu-latest' )  || (github.event_name != 'pull_request') || (github.event_name == 'pull_request' && github.event.action == 'review_requested')}}
       shell: bash
       run: nox -v -s tests_brevitas_cpu-${{ matrix.python_version }}\(${{ matrix.jit_status }}\,\ pytorch_${{ matrix.pytorch_version }}\)
-
-

.github/workflows/rebase.yml

@@ -24,4 +24,3 @@ jobs:
     steps:
       - name: Always run
         run: echo "This job is used to prevent the workflow to fail when all other jobs are skipped."
-

.github/workflows/utils.py

@@ -1,11 +1,10 @@
+from collections import OrderedDict as od
 from functools import reduce
 from string import Template
 from textwrap import indent
-from collections import OrderedDict as od
 
 import yaml
 
-
 NIX_NEWLINE = '\n'
 # whitespaces to indent generated portions of output yaml
 STEP_INDENT = 4
@@ -66,7 +65,7 @@ def gen_yaml(self, base_template_path, output_path):
                 Action.list_of_dicts_str(self.exclude_list, False, True, True), EXCLUDE_INDENT * ' ')
         else:
             d['exclude'] = ''
-        
+
         d['strategy'] = indent(Action.dict_str(self.strategy_params, False, False), 6 * ' ')
         template = CustomTemplate(open(base_template_path).read())
         generated_file = template.substitute(d)
@@ -76,4 +75,4 @@ def gen_yaml(self, base_template_path, output_path):
 
 
 def combine_od_list(od_list):
-    return od(reduce(lambda l1, l2: l1 + l2, list(map(lambda d: list(d.items()), od_list))))
+    return od(reduce(lambda l1, l2: l1 + l2, list(map(lambda d: list(d.items()), od_list))))

.gitignore

@@ -27,7 +27,7 @@ data
 *.onnx
 
 # Commented out files
-*.bak 
+*.bak
 
 # Installer logs
 pip-log.txt

.pre-commit-config.yaml

@@ -0,0 +1,31 @@
+
+default_language_version:
+    python: python3
+
+default_stages: [commit]
+
+exclude: 'docs'
+
+repos:
+- repo: https://github.com/pre-commit/pre-commit-hooks
+  rev: v4.0.1
+  hooks:
+  - id: trailing-whitespace
+  - id: check-added-large-files
+    args: [--maxkb=5000]
+  - id: check-ast
+  - id: check-json
+  - id: check-merge-conflict
+  - id: check-xml
+  - id: check-yaml
+    args: [--allow-multiple-documents]
+  - id: debug-statements
+  - id: end-of-file-fixer
+  - id: requirements-txt-fixer
+  - id: mixed-line-ending
+
+-   repo: https://github.com/pycqa/isort
+    rev: 5.10.1
+    hooks:
+    -   id: isort
+        name: isort (python)

LICENSE

@@ -18,4 +18,4 @@ mechanisms:
    `LICENSE` file at the top containing the specific license and restrictions
    which apply to that software, or
 2) It will contain specific license and restriction terms at the top of every
-   file.
+   file.

README.md

@@ -10,7 +10,7 @@ Brevitas is a PyTorch library for neural network quantization, with a focus on *
 
 **Please note that Brevitas is a research project and not an official Xilinx product.**
 
-If you like this project please consider ⭐ this repo, as it is the simplest and best way to support it. 
+If you like this project please consider ⭐ this repo, as it is the simplest and best way to support it.
 
 If you have issues, comments, or are just looking for advices on training quantized neural networks, open an issue or a discussion.
 
@@ -37,7 +37,7 @@ If you adopt Brevitas in your work, please cite it as:
 - *2021/10/29* - Release version 0.7.0, see the [release notes](https://github.com/Xilinx/brevitas/releases/tag/v0.7.0).
 - *2021/06/04* - Release version 0.6.0, see the [release notes](https://github.com/Xilinx/brevitas/releases/tag/v0.6.0).
 - *2021/05/24* - Release version 0.5.1, fix a bunch of minor issues. See [release notes](https://github.com/Xilinx/brevitas/releases/tag/v0.5.1).
-- *2021/05/06* - Release version 0.5.0, see the [release notes](https://github.com/Xilinx/brevitas/releases/tag/v0.5.0). 
+- *2021/05/06* - Release version 0.5.0, see the [release notes](https://github.com/Xilinx/brevitas/releases/tag/v0.5.0).
 - *2021/03/15* - Release version 0.4.0, add support for \_\_torch_function\_\_ to QuantTensor.
 - *2021/03/04* - Release version 0.3.1, fix bug w/ act initialization from statistics w/ IGNORE_MISSING_KEYS=1.
 - *2021/03/01* - Release version 0.3.0, implements enum and shape solvers within extended dependency injectors. This allows declarative quantizers to be self-contained.
@@ -61,9 +61,3 @@ pip install brevitas
 ## Getting started
 
 Check out available info at https://xilinx.github.io/brevitas/getting_started .
-
-
-
-
-
-

docsrc/source/about.rst

@@ -20,4 +20,4 @@ Cite as
      publisher    = {Zenodo},
      doi          = {10.5281/zenodo.3333552},
      url          = {https://doi.org/10.5281/zenodo.3333552}
-   }
+   }

docsrc/source/api_reference/brevitas.function.rst

@@ -27,5 +27,3 @@ brevitas.function.shape module
    :members:
    :undoc-members:
    :show-inheritance:
-
-

docsrc/source/architecture.rst

@@ -222,4 +222,4 @@ TODO
 Losses
 ------
 
-TODO
+TODO

docsrc/source/conf.py

@@ -16,6 +16,7 @@
 
 import os
 import sys
+
 import brevitas
 
 sys.path.insert(0, os.path.abspath(brevitas.__file__))
@@ -93,4 +94,4 @@
     "numpy": ("http://docs.scipy.org/doc/numpy/", None),
     "python": ("https://docs.python.org/", None),
     "torch": ("https://pytorch.org/docs/master/", None),
-}
+}

docsrc/source/faq.rst

@@ -5,12 +5,12 @@ F.A.Q.
 **Q: Pytorch supports quantization-aware training. Why should I use
 Brevitas?**
 
-**A:** Quantization in Pytorch is designed to target specific backends 
-supported by PyTorch itself (currently FBGEMM and qnnpack). 
+**A:** Quantization in Pytorch is designed to target specific backends
+supported by PyTorch itself (currently FBGEMM and qnnpack).
 
 Brevitas is designed as a platform to implement novel quantization
 algorithms to target a variety of hardware backends adhering to a loose
-set of assumptions (i.e. uniform affine quantization). 
+set of assumptions (i.e. uniform affine quantization).
 
 **Q: How can I train X/Y and run it on hardware W/Z? I can't find any
 documentation.**
@@ -41,11 +41,11 @@ make my model faster at inference time. What I am doing wrong?**
 **A:** Brevitas is concerned with modelling a reduced precision
 data-path, it does not provide inference-time acceleration on its own.
 To achieve acceleration, you should export your Brevitas model to a
-downstream toolchain / backend. See the 
+downstream toolchain / backend. See the
 
 **Q: My (C/G/T)PU supports float16 / bfloat16 / bfloat19 training. Can I
 use it to train with Brevitas?**
 
 **A:** Datatypes outside of float32 at training time have not been tested. That includes training on TPU / Pytorch-XLA.
 Do the math in terms of which reduced-precision integers can reasonably fit in a reduced-precision
-floating-point format at training time, and use at your own risk.
+floating-point format at training time, and use at your own risk.

docsrc/source/getting_started.rst

@@ -62,17 +62,17 @@ The result is the following:
    # ... training ...
 
 A neural network with 4 bits weights and floating-point
-activations can provide an advantage in terms of model storage, 
-but it doesn't provide any advantage in terms of compute, 
-as the weights need to be converted to float at runtime first. 
-In order to make more practical, we want to quantize activations too. 
+activations can provide an advantage in terms of model storage,
+but it doesn't provide any advantage in terms of compute,
+as the weights need to be converted to float at runtime first.
+In order to make more practical, we want to quantize activations too.
 
 Weights and activations quantization, float biases
 --------------------------------------------------
 
-We now quantize both weights and activations to 4 bits, while keeping biases in floating-point. 
+We now quantize both weights and activations to 4 bits, while keeping biases in floating-point.
 In order to do so, we replace ``torch.nn.ReLU`` with
-``brevitas.nn.QuantReLU``, specifying ``bit_width=4``. 
+``brevitas.nn.QuantReLU``, specifying ``bit_width=4``.
 Additionally, in order to quantize the very first input, we introduce a
 ``brevitas.nn.QuantIdentity`` at the beginning of the network. The end
 result is the following:
@@ -166,7 +166,7 @@ Weights, activations, biases quantization
 
 
 Compared to the previous scenario:
-- We now set ``return_quant_tensor=True`` in every quantized activations to propagate a ``QuantTensor`` to the next layer. This informs each ``QuantLinear`` or ``QuantConv2d`` of how the input passed in has been quantized. 
+- We now set ``return_quant_tensor=True`` in every quantized activations to propagate a ``QuantTensor`` to the next layer. This informs each ``QuantLinear`` or ``QuantConv2d`` of how the input passed in has been quantized.
 - A ``QuantTensor`` is just a tensor-like data structure providing metadata about how a tensor has been quantized, similar to a `torch.qint` dtype, but training friendly. Setting ``return_quant_tensor=True`` does not affect the way quantization is performed, it only changes the way the output is represented.
 - We enable bias quantization by setting the `Int32Bias` quantizer. What it does is to perform bias quantization with ```bias_scale = input_scale * weight_scale``, as it commonly done across inference toolchains. This is why we have to set ``return_quant_tensor=True``: each layer with ``Int32Bias`` can read the input scale from the ``QuantTensor`` passed in and use for bias quantization.
 - ``torch`` operations that are algorithmically invariant to quantization, such as `F.max_pool2d`, can propagate QuantTensor through them without extra changes.
@@ -175,8 +175,8 @@ Export to ONNX
 --------------
 
 Brevitas does not perform any low-precision acceleration on its own. For that to happen, the model need to be exported first to an inference toolchain through some intermediate representation like ONNX.
-One popular way to represent 8-bit quantization within ONNX is through the `QDQ format <https://onnxruntime.ai/docs/performance/quantization.html#onnx-quantization-representation-format>`_. 
-Brevitas extends *QDQ* to **QCDQ**, inserting a `Clip` node to represent quantization to *<= 8 bits*. We can then export all previous defined model to *QCDQ*. 
+One popular way to represent 8-bit quantization within ONNX is through the `QDQ format <https://onnxruntime.ai/docs/performance/quantization.html#onnx-quantization-representation-format>`_.
+Brevitas extends *QDQ* to **QCDQ**, inserting a `Clip` node to represent quantization to *<= 8 bits*. We can then export all previous defined model to *QCDQ*.
 The interface of the export function matches the `torch.onnx.export` function, and accepts all its kwargs:
 
 .. code-block:: python
@@ -196,4 +196,4 @@ The interface of the export function matches the `torch.onnx.export` function, a
 Where to go from here
 ---------------------
 
-Check out the :doc:`Tutorials section </tutorials/index>` for more information on things like ONNX export, quantized recurrent layers, quantizers, or a more detailed overview of the library in the TVMCon tutorial. 
+Check out the :doc:`Tutorials section </tutorials/index>` for more information on things like ONNX export, quantized recurrent layers, quantizers, or a more detailed overview of the library in the TVMCon tutorial.

docsrc/source/index.rst

@@ -27,4 +27,4 @@ It provides a platform both for researchers interested in implementing new quant
 Brevitas supports a super-set of quantization schemes implemented across various frameworks and compilers under a single unified API.
 For certain combinations of layers and types of of quantization inference acceleration is supported by exporting to *FINN*, *onnxruntime* or *Pytorch*'s own quantized operators.
 
-Brevitas has been successfully adopted both in various research projects as well as in large-scale commercial deployments targeting CPUs, GPUs, and custom accelerators running on Xilinx FPGAs. The general quantization style implemented is affine quantization, with a focus on uniform quantization. Non-uniform quantization is currently not supported out-of-the-box.
+Brevitas has been successfully adopted both in various research projects as well as in large-scale commercial deployments targeting CPUs, GPUs, and custom accelerators running on Xilinx FPGAs. The general quantization style implemented is affine quantization, with a focus on uniform quantization. Non-uniform quantization is currently not supported out-of-the-box.