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

fix q4_0_8_8 format for corrupted tokens issue

Browse source
This commit is contained in:
EC2 Default User 2024-11-07 05:57:52 +00:00
parent 5c333e0140
commit 26f3174c8f
1 changed files with 2 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml/src/ggml-cpu.c
View file

@ -409,6 +409,8 @@ static const struct ggml_type_traits_cpu type_traits_cpu[GGML_TYPE_COUNT] = {
.gemm = ggml_gemm_q4_0_4x8_q8_0, .gemm = ggml_gemm_q4_0_4x8_q8_0,
}, },
[GGML_TYPE_Q4_0_8_8] = { [GGML_TYPE_Q4_0_8_8] = {
.vec_dot = NULL,
.vec_dot_type = GGML_TYPE_Q8_0,
.nrows = 1, .nrows = 1,
.ncols = 8, .ncols = 8,
.gemv = ggml_gemv_q4_0_8x8_q8_0, .gemv = ggml_gemv_q4_0_8x8_q8_0,