TurboQuant: head_dim as a multiple of 128, three fixes, test coverage, fused kernel - #2

TurboQuant: head_dim as a multiple of 128, three fixes, test coverage, fused kernel#2

Pascal-SAPUI5 merged 18 commits into

masterPascal-SAPUI5/llama.cpp-turboquant:masterfrom

feat/turboquant-head-dim-multiples-of-128Pascal-SAPUI5/llama.cpp-turboquant:feat/turboquant-head-dim-multiples-of-128Copy head branch name to clipboard