cmdr2
Support pure float16 add/sub/mul/div operations in the CUDA (and CPU) backend (ggml/1121)
2b94a24