Qwen3-TTS OpenVINO port targets Intel hardware

Hot take: this reads more like an engineering notebook than a standard model port, and that makes it valuable for anyone trying to understand how OpenVINO optimization actually works in practice.

• –Strong educational angle: it explains the conversion path from PyTorch module logic to OpenVINO-friendly execution.
• –Clear technical focus: the hard parts are stateful KV cache behavior, kernel fusion, and letting the compiler choose good kernels.
• –Practical scope: the implementation targets 1.7B on CPU/GPU only, so it is useful but not yet broadly packaged.
• –Limited benchmarking: the author explicitly says there are no benchmarks in the post, so performance claims are unproven here.
• –Good signal for contributors: NPU support is explicitly open for PRs, which suggests the repo is meant to evolve.