Streaming FR→EN speech-to-speech: whisper-large-v3 → Qwen2.5-7B gist → Kokoro-82M. Serves /v1/predict + /predict. See server.py / gist_model.py.
