serizba / cppflow

Run TensorFlow models in C++ without installation and without Bazel
https://serizba.github.io/cppflow/
MIT License
781 stars 177 forks source link

Added method to create tensors and models from data in memory. #264

Open 8-byte opened 1 month ago

8-byte commented 1 month ago

Added static member to tensor class to allows the creation of a tensor directly from an existing buffer without having to create a copy as a std:vector. Specially useful when the input already exists in raw format in memory and has a significant footprint. Also added member to model class to allow building a model from a buffer already in memory that's been loaded with the content of a saved model file. While this use case may appear obscure, there are cases where a stand-alone application needs to be distributed with encrypted models that are only decrypted in memory. Note: the fix on the status member was an oversight while merging my old codebase with the new fork of cppflow:master. The new fixed constructor model compiles and runs as expected.