llama.cpp
Use BLAS to implement ggml_compute_forward_out_prod_f32 for matrix src0, src1 (finetuning speedup ~5x).
#4079
Merged
Merged