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

llama : check for 256 divisibility for IQ2_XS, IQ2_XXS (#4950)

Browse source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow 2024-01-15 10:09:38 +02:00 • committed by GitHub
parent 4a3156de2f
commit 2faaef3979
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

5
llama.cpp
View file

@ -8559,7 +8559,8 @@ static ggml_type get_k_quant_type(quantize_state_internal & qs, ggml_type new_ty
//}
bool convert_incompatible_tensor = false;
if (new_type == GGML_TYPE_Q2_K || new_type == GGML_TYPE_Q3_K || new_type == GGML_TYPE_Q4_K ||
new_type == GGML_TYPE_Q5_K || new_type == GGML_TYPE_Q6_K) {
new_type == GGML_TYPE_Q5_K || new_type == GGML_TYPE_Q6_K ||
new_type == GGML_TYPE_IQ2_XS || new_type == GGML_TYPE_IQ2_XXS) {
int nx = tensor->ne[0];
int ny = tensor->ne[1];
if (nx % QK_K != 0) {
@ -8571,6 +8572,8 @@ static ggml_type get_k_quant_type(quantize_state_internal & qs, ggml_type new_ty
}
if (convert_incompatible_tensor) {
switch (new_type) {
case GGML_TYPE_IQ2_XXS:
case GGML_TYPE_IQ2_XS:
case GGML_TYPE_Q2_K: new_type = GGML_TYPE_Q4_0; break;
case GGML_TYPE_Q3_K: new_type = GGML_TYPE_Q4_1; break;
case GGML_TYPE_Q4_K: new_type = GGML_TYPE_Q5_0; break;