Retire the ggml_mul_mat() for transposed src0
- It can always be made contiguous with ggml_cpy() - The code is now simplified - The results are deterministic in respect to num threads
This commit is contained in:
parent
502a400192
commit
1e39d2bf77