ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-11 22:40:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

I somehow broke iq2_k on Metal? - fix dequantize

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-10-13 13:47:46 +03:00
parent 910a134094
commit 332049da85
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-metal.metal
View File

@@ -7705,10 +7705,10 @@ template <typename type4x4>
void dequantize_iq2_k(device const block_iq2_k * xb, short il, thread type4x4 & reg) {
// il is 0...15 for QK_K = 256
device const uint32_t * q32 = (device const uint32_t *)xb->qs + 8*(il/8) + 4*(il&1);
half d = xb->d * (2*((xb->scales[il/2] >> 4*(il&1)) & 0xf) - 15);
half d = xb->d * (((xb->scales[il/2] >> 4*(il&1)) & 0xf) - 8);
constant int8_t * int_values = iq2nl_values + 4*((xb->extra >> il) & 1);
half4 values = { d * int_values[0], d * int_values[1], d * int_values[2], d * int_values[3] };
constant half4 * half_values = (constant half4 *)kvalues_iq2k_h;
half4 values = half_values[(xb->extra >> il) & 1] * d;
const int shift = 2*((il%8)/2);
uint32_t aux32;