QAT of RNN(LSTM)

Swaraj_Badhei · June 27, 2023, 5:58am

There exists a commented section in tfmot for QAT of RNN layers.

tensorflow/model-optimization/blob/master/tensorflow_model_optimization/python/core/quantization/keras/default_8bit/default_8bit_quantize_registry.py

# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Quantization registry which specifies how layers should be quantized."""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

This file has been truncated. show original

class Default8BitQuantizeRegistry
Line : 170.

The implementation seems semi-developed.
I want to extend this and make it work for LSTMs. However, I am not sure where and how to start.
I would really appreciate any starting points on the same.

Thanks a lot in advance.

chunduriv · June 27, 2023, 7:21am

@Swaraj_Badhei,

Thank you for showing interest to contribute. Before you contribute source code to a TensorFlow project, please review the CONTRIBUTING.md file in the GitHub repo of the project. All code contributors are required to sign a Contributor License Agreement (CLA).

Contribution workflow

Code contributions—bug fixes, new development, test improvement—all follow a GitHub-centered workflow. To participate in TensorFlow development, set up a GitHub account. Then:

Fork the repo you plan to work on. Go to the project repo page and use the Fork button. This will create a copy of the repo, under your username. (For more details on how to fork a repository see this guide.)
Clone down the repo to your local system.$ git clone git@github.com:your-user-name/project-name.git
Create a new branch to hold your work.$ git checkout -b new-branch-name
Work on your new code. Write and run tests.
Commit your changes.$ git add -A``$ git commit -m "commit message here"
Push your changes to your GitHub repo.$ git push origin branch-name
Open a Pull Request (PR). Go to the original project repo on GitHub. There will be a message about your recently pushed branch, asking if you would like to open a pull request. Follow the prompts, compare across repositories, and submit the PR. This will send an email to the committers. You may want to consider sending an email to the mailing list for more visibility. (For more details, see the GitHub guide on PRs.
Maintainers and other contributors will review your PR. Please participate in the conversation, and try to make any requested changes. Once the PR is approved, the code will be merged.

For more details please refer to Contribute to the TensorFlow code