vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix typo that makes several tests fail

Browse source
This commit is contained in:
Iwan Kawrakow 2024-02-20 09:21:11 +02:00
parent 3fc455584f
commit b376bbb21d
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml-quants.c
View file

@ -9511,8 +9511,8 @@ void ggml_vec_dot_iq4_nl_q8_0(int n, float * restrict s, size_t bs, const void *
q4b.val[2] = vqtbl1q_s8(values, vandq_u8(q4bits.val[1], m4b));
q4b.val[3] = vqtbl1q_s8(values, vshrq_n_u8(q4bits.val[1], 4));
prod_1 = ggml_vdotq_s32(vdotq_s32(vdupq_n_s8(0), q4b.val[0], q8b.val[0]), q4b.val[1], q8b.val[1]);
prod_2 = ggml_vdotq_s32(vdotq_s32(vdupq_n_s8(0), q4b.val[2], q8b.val[2]), q4b.val[3], q8b.val[3]);
prod_1 = ggml_vdotq_s32(vdotq_s32(vdupq_n_s32(0), q4b.val[0], q8b.val[0]), q4b.val[1], q8b.val[1]);
prod_2 = ggml_vdotq_s32(vdotq_s32(vdupq_n_s32(0), q4b.val[2], q8b.val[2]), q4b.val[3], q8b.val[3]);
sumf += (float)x[ib+0].d * (float)y[ib+0].d * vaddvq_s32(prod_1) + (float)x[ib+1].d * (float)y[ib+1].d * vaddvq_s32(prod_2);