alexlnkp/Remi-CPP

Remi-CPP

Getting started

Prerequisites

• C/C++ compiler {gcc, clang}
• CMake 3.29 or higher
• CTranslate2 (refer to CTranslate2 installation)

Building

1. Recursively clone this repository git clone --recursive https://github.com/alexlnkp/remi-cpp.git
2. Enter the cloned repository.
3. Run chmod +x autobuild to make the build script script executable.
4. Run ./autobuild to build with a single click.

Usage

1. Make sure your model is compatible with CTranslate2. If it isn't - use the CTConverter script
   • Refer to CTranslate2 documentation for list of conversion-compatible models.
2. Tweak the config.json file accordingly to modify the inference parameters.
3. So far you need to change source code to modify prompt, but soon (when i fix the broken generation) you won't have to.
4. Enjoy... I guess? Most of the results so far are stuff like this:

## Response:
Assistant: 2147453983566521C*24...722-2262609322,CM2x3XP (5241)41D832MC

177333323C7
-5476193

License

This project is licensed under the MIT License - see the LICENSE file for details.

TODO

1. A working inference
2. Adapter loader
3. A working fine-tuner

Credit

• Hedley made by nemequ
• CTranslate2 made by OpenNMT
• JSON library made by Nlohmann

Misc

NLohmann's JSON library installation

1. Clone Nlohmann/JSON repository git clone https://github.com/nlohmann/json.git
2. Enter the cloned repository.
3. Run cmake -B build
4. Run cmake --install build

CTranslate2 installation

1. Recursively clone CTranslate2 repository git clone --recursive https://github.com/OpenNMT/CTranslate2.git
2. Enter the cloned repository.
3. Run cmake -B build
4. Run cmake --build build
5. Run cmake --install build

CTConverter script

1. Create a virtual environment using python -m venv .venv
2. Activate the virtual environment using source .venv/bin/activate (Linux) or . .venv/Scripts/activate (Windows)
3. Install requirements using pip install -r scripts/converter.req
4. Run the script using python scripts/ctconverter.py