snadampal EC2 Default User commited on
Commit
4700b48
·
1 Parent(s): f58e658

fix q4_0_8_8 format for corrupted tokens issue (llama/10198)

Browse files

Co-authored-by: EC2 Default User <ec2-user@ip-172-31-62-167.us-west-2.compute.internal>

Files changed (1) hide show
  1. ggml/src/ggml-cpu.c +2 -0
ggml/src/ggml-cpu.c CHANGED
@@ -409,6 +409,8 @@ static const struct ggml_type_traits_cpu type_traits_cpu[GGML_TYPE_COUNT] = {
409
  .gemm = ggml_gemm_q4_0_4x8_q8_0,
410
  },
411
  [GGML_TYPE_Q4_0_8_8] = {
 
 
412
  .nrows = 1,
413
  .ncols = 8,
414
  .gemv = ggml_gemv_q4_0_8x8_q8_0,
 
409
  .gemm = ggml_gemm_q4_0_4x8_q8_0,
410
  },
411
  [GGML_TYPE_Q4_0_8_8] = {
412
+ .vec_dot = NULL,
413
+ .vec_dot_type = GGML_TYPE_Q8_0,
414
  .nrows = 1,
415
  .ncols = 8,
416
  .gemv = ggml_gemv_q4_0_8x8_q8_0,