wubet/amharic-fairseq

# amharic-fairseq The Amharic-fairseq framework originates directly from the fairseq toolkit, taken from commit version a8f28ecb63ee01c33ea9f6986102136743d47ec2. This framework is customized for bidirectional Amharic-English sequence translation and utilizes attention mechanisms, moving away from the previously prevalent recurrent neural network (RNN) frameworks. ### Requirements and Installation: PyTorch version >= 1.13.0 \ Python version >= 3.6 Cloning and Building the Repository: Repo cloning: ```commandline git clone github.com ``` Change directory to the clone project ```commandline cd amharic-fairseq ``` Install required dependency ```commandline pip3 install sacremoses cd exampes/translation pip3 install github.com pip3 install github.com ``` get back to the main project ```commandline cd ../../ ``` Installs the build module, a modern Python package builder ```commandline pip3 install build ``` Builds the package, generating distribution archives (wheel and source) in the dist directory ```commandline python3 -m build ``` Installs the package in editable mode (also known as development mode), allowing changes to the code to be immediately reflected ```commandline pip3 install --editable . ``` Compiles and builds any extension modules (e.g., C extensions) specified in setup.py directly in the source directory ```commandline python3 setup.py build_ext –inplace ``` ### Data Preparation Clone the English-Amharic corpus. ```commandline Git clone github.com ``` Process the corpus data for training and evaluation. Typically, processing encompasses various crucial stages to adeptly navigate the intricacies of language. Such stages encompass breaking down words into subwords or tokens, mapping these tokens to a specific vocabulary, and incorporating special tokens, all aimed at enhancing the model's proficie …