Windows Packaging (uv + PyInstaller)

This guide builds a single-file .exe for the Trans-for-Doctors CLI using uv and PyInstaller.

Prerequisites

Windows 10/11
Python 3.10+ installed
uv installed (https://docs.astral.sh/uv/)
Adequate disk space (bundling torch/transformers + model can be several GB)

Install dependencies with uv

uv pip install .
uv pip install .[llm]
uv pip install pyinstaller

If you use CUDA builds of torch, ensure matching CUDA runtime is installed.

Build single-file exe

Build the CLI app/main.py as a single executable named TransForDoctors.exe.

uv run pyinstaller \
  --onefile \
  --name TransForDoctors \
  app/main.py \
  --add-data "config.json;." \
  --add-data "tokenizer_config.json;." \
  --add-data "preprocessor_config.json;." \
  --add-data "normalizer.json;." \
  --add-data "special_tokens_map.json;." \
  --add-data "added_tokens.json;." \
  --add-data "merges.txt;." \
  --add-data "vocab.json;." \
  --add-data "medical_terms.txt;."

Notes:

--add-data pairs use SRC;DEST with ; on Windows.
Including model.safetensors will produce a very large binary; you can omit it and place the model files next to the .exe under the same directory if preferred.

Running

Place TransForDoctors.exe alongside the model files (if not embedded) and run:

./TransForDoctors.exe \
  --audio path\\to\\dictation.wav \
  --model . \
  --terms medical_terms.txt \
  --llm \
  --openai-key %OPENAI_API_KEY% \
  --save-original --save-corrected --generate-report

Troubleshooting

Missing CUDA DLLs: use CPU (--device cpu) or install matching CUDA runtime.
Large binary size: omit --add-data model.safetensors and keep model files next to the exe.
OpenAI key: set environment variable OPENAI_API_KEY or pass --openai-key.