The main ggml.c file currently contains ~20,000 lines of code, which makes navigation, debugging, and code review difficult. Similar large files include ggml-cuda.cu and ggml-vulkan.cpp.
We kindly request splitting these files by functionality — such as separating tensor operations, quantization logic, and backend-specific code into dedicated modules. This would significantly improve code organization, reduce merge conflicts, and most importantly, make it easier to port and maintain these modules in dependent packages (e.g., ggmlR).
Smaller, focused files are much easier to track, patch, and integrate into downstream projects while maintaining version compatibility.
The main ggml.c file currently contains ~20,000 lines of code, which makes navigation, debugging, and code review difficult. Similar large files include ggml-cuda.cu and ggml-vulkan.cpp.
We kindly request splitting these files by functionality — such as separating tensor operations, quantization logic, and backend-specific code into dedicated modules. This would significantly improve code organization, reduce merge conflicts, and most importantly, make it easier to port and maintain these modules in dependent packages (e.g., ggmlR).
Smaller, focused files are much easier to track, patch, and integrate into downstream projects while maintaining version compatibility.