ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

vulkan: initial support for IQ4_XS quantization (#11501)

Browse source
This commit is contained in:
Rémy O 2025-02-06 07:09:59 +01:00 • committed by GitHub
parent 1b598b3058
commit 8a7e3bf17a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
13 changed files with 169 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

38
ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.comp
View file

@ -304,6 +304,42 @@ vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
}
#endif
#if defined(DATA_A_IQ4_XS)
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
const uint ib32 = iqs / 32;
const uint iq = 16 * ib32 + (iqs % 16);
const uint sl = (data_a[a_offset + ib].scales_l[ib32/2] >> (4 * (ib32 & 1))) & 0xF;
const uint sh = (data_a[a_offset + ib].scales_h >> (2 * ib32)) & 3;
const uint qshift = (iqs & 16) >> 2;
u8vec2 qs = u8vec2(data_a[a_offset + ib].qs[iq], data_a[a_offset + ib].qs[iq + 1]);
qs = (qs >> qshift) & uint8_t(0xF);
const float dl = float(int(sl | (sh << 4)) - 32);
return dl * vec2(kvalues_iq4nl[qs.x], kvalues_iq4nl[qs.y]);
}
vec4 dequantize4(uint ib, uint iqs, uint a_offset) {
const uint ib32 = iqs / 32;
const uint iq = 16 * ib32 + (iqs % 16);
const uint sl = (data_a[a_offset + ib].scales_l[ib32/2] >> (4 * (ib32 & 1))) & 0xF;
const uint sh = (data_a[a_offset + ib].scales_h >> (2 * ib32)) & 3;
const uint qshift = (iqs & 16) >> 2;
u8vec4 qs = u8vec4(
data_a[a_offset + ib].qs[iq + 0],
data_a[a_offset + ib].qs[iq + 1],
data_a[a_offset + ib].qs[iq + 2],
data_a[a_offset + ib].qs[iq + 3]
);
qs = (qs >> qshift) & uint8_t(0xF);
const float dl = float(int(sl | (sh << 4)) - 32);
return dl * vec4(
kvalues_iq4nl[qs.x], kvalues_iq4nl[qs.y],
kvalues_iq4nl[qs.z], kvalues_iq4nl[qs.w]);
}
#endif
#if defined(DATA_A_IQ4_NL)
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
const uint vui = uint(data_a[a_offset + ib].qs[iqs]);
@ -321,7 +357,7 @@ vec2 get_dm(uint ib, uint a_offset) {
}
#endif
#if defined(DATA_A_Q4_0) || defined(DATA_A_Q5_0) || defined(DATA_A_Q8_0) || defined(DATA_A_IQ2_XXS) || defined(DATA_A_IQ2_XS) || defined(DATA_A_IQ2_S) || defined(DATA_A_IQ3_XXS) || defined(DATA_A_IQ3_S) || defined(DATA_A_IQ4_NL)
#if defined(DATA_A_Q4_0) || defined(DATA_A_Q5_0) || defined(DATA_A_Q8_0) || defined(DATA_A_IQ2_XXS) || defined(DATA_A_IQ2_XS) || defined(DATA_A_IQ2_S) || defined(DATA_A_IQ3_XXS) || defined(DATA_A_IQ3_S) || defined(DATA_A_IQ4_XS) || defined(DATA_A_IQ4_NL)
vec2 get_dm(uint ib, uint a_offset) {
return vec2(float(data_a[a_offset + ib].d), 0);
}