During January 2025, Oudomphe Phare developed initial Vulkan-based IQ2 and IQ3 quantization support for the Mintplex-Labs/whisper.cpp and ggml-org/llama.cpp repositories. Leveraging C++, GLSL, and GPU computing expertise, Oudomphe implemented new shader pipelines and dequantization routines, enabling faster and more memory-efficient inference on Vulkan-enabled hardware. The work included code alignment and optimizations for Q3_K quantization, ensuring consistent APIs and integration points across both projects. By establishing parallel quantization support paths, Oudomphe expanded hardware compatibility and laid the groundwork for future performance improvements, addressing the need for efficient quantized model inference in performance-sensitive deployment scenarios.