tf-sha256 is a Python library for computing SHA256 with TensorFlow.
SHA256 is a common cryptographic hash function. This library provides an implementation of SHA256 with TensorFlow, which means it enables us to potentially design loss functions with it to train ML models.
Use the package manager poetry to install tf-sha256.
poetry installNote I haven't published this package to Pypi since in my own experience, the compatibility of TensorFlow and CUDA varies, so it might be easier for you to install this directly from the repo, or even copy-paste/submodule the files to your project, such that you can tweak things for your setup.
Please refer to the unit tests for a more detailed usage of the library. Note this library has not been used in any production workload, despite reaching parity with non-tf implementation of SHA256 (as shown by the unit tests).
A well-known use case for SHA256 is Bitcoin, so in tf_btcminer we present a hypothetical use of this library in training a model to optimize the Bitcoin mining loss (which is based on the nonce and the coinbase value the model predict for a given block template). Again, you can find examples of the loss function in the tests directory.
To make sure it works, I have tried trainning a minimal ML model to optimize this loss (with 1 RTX 3060). So far, I haven't managed to accelerate the training enough to make meaningful progress. Therefore, I think there's still plenty of space for compute optimization, or a larger compute cluster is needed.
Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Please make sure to update tests as appropriate.
Consider sponsoring the project if this turns out useful☕