Transcription deployment

VoicePad currently exposes one production deployment:

parakeet-v3.transformers-fp16-cuda

It combines NVIDIA’s official Parakeet TDT 0.6B v3 safetensors, Transformers, PyTorch FP16 CUDA, immutable artifact hashes, strict device admission, and the bounded chunk policy.

There is no model dropdown because unvalidated model/runtime combinations are not user settings. A future deployment must define its own artifacts, adapter, capabilities, resource profile, and quality evidence before it appears here.