llama.rn
llama.rn copied to clipboard
feat(ios): try enable Metal for older iOS devices
Context: https://github.com/ggerganov/llama.cpp/discussions/2550
Currently:
- only replaced methods related to q2_k / q4_k / q6_k models
- not able to get the correct result, need more time to understand
Probably won't merge even if it succeeds because it would complicate the code base. Just curious about the performance of the A12Z iPad Pro here.