FluidVoice Windows 0.0.5 beta release.
- Greatly accelerates Fluid Intelligence Boost decoding on supported NVIDIA GPUs while preserving exact greedy output.
- Adds prompt-prefix caching and configurable MTP draft depth for faster repeated corrections and longer generations.
- Reduces perceived dictation latency by decoupling overlay transitions from text insertion.
- Adds a verified on-demand CUDA optimization overlay with CPU/Vulkan fallback and immediate CPU-to-CUDA switching after setup.