jastin0012/Swahili_gpt2

Swahili GPT-2 language model with a C-based inference server powered by ONNX Runtime and Mongoose, exposing an OpenAI-compatible chat completions API ## Swahili GPT-2 Submission This project implements a Swahili GPT-2-style inference pipeline in C using ONNX Runtime, with a lightweight HTTP server stub built on Mongoose. It demonstrates: - training support via `train_swahili.py` and whole notebook in `train.ipynb` - tokenizer export and model export workflows - tokenization and inference plumbing in C - a minimal OpenAI-style `/v1/chat/completions` server stub --- ## Repository Layout - `src/` — core C sources and headers for the inference engine and server - `model/` — model artifacts and tokenizer files - `scripts/` — helper utilities for model file setup and build support - `tools/` — testing and inference utilities - `python_server/` — Python-based server/demo support --- ## IMPORTANT: Model Files (Download Required) The large model files are not guaranteed to be included in the GitHub repository due to size limits. Download them manually and place them in `model/`. - **model.onnx** drive.google.com - **model.safetensors** drive.google.com Place both files inside: ```powershell C:\Users\YourUser\Desktop\Swahili_gpt2\model ``` If `model.onnx` or `model.safetensors` are missing, the C inference and server builds will not work correctly. --- ## Build Requirements - Windows with GCC or MSYS2/Mingw-w64 installed - `mongoose.c` and `mongoose.h` available in `externals/mongoose/` or `third_party/mongoose/` Optional for real inference: - ONNX Runtime C SDK for Windows - `ORT_INCLUDE` and `ORT_LIB` environment variables set to the ONNX Runtime SDK paths Example PowerShell settings: ```powershell $env:ORT_INCLUDE = 'C:\externals\onnxruntime\include' $env:ORT_LIB = 'C:\externals\onnxruntime\lib' ``` If `ORT_INCLUDE`/`ORT_LIB` are not set, `build.ps1` will still compile a stub C server and engine. The stub s …