ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-27 00:24:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

iq4_nl

Browse Source 
137 t/s -> 293 t/s. iq4_nl is at 251 t/s.
This commit is contained in:
Iwan Kawrakow
2025-06-18 10:33:46 +03:00
parent 516b1d54f1
commit 0788a71fcf
2 changed files with 15 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
ggml/src/iqk/iqk_gemm_legacy_quants.cpp
View File

@@ -551,6 +551,14 @@ struct IQ4_NL_Dequantizer {
}
};
struct IQ4_NL0_Dequantizer {
Dequantizer4bit b4;
const __m256i values = load_iq4k_values_256();
inline __m256i dequant(const block_iq4_nl * x) const {
return _mm256_shuffle_epi8(values, b4.dequant(x->qs));
}
};
struct Q4_1_Dequantizer {
Dequantizer4bit b4;
inline __m256i dequant(const block_q4_1 * x) const {
@@ -1792,10 +1800,11 @@ template <typename Dequantizer> void set_functions(std::array<mul_mat_t, IQK_MAX
bool iqk_convert_legacy_quants_q8_r8(int type, int n, const void * vx, size_t bx, void * vy, int nrc_x) {
switch (type) {
case GGML_TYPE_Q4_0: iqk_convert_qX_q80_r8<block_q4_0, Q4_0_Dequantizer>(n, vx, bx, vy, nrc_x); break;
case GGML_TYPE_Q5_0: iqk_convert_qX_q80_r8<block_q5_0, Q5_0_Dequantizer>(n, vx, bx, vy, nrc_x); break;
case GGML_TYPE_Q6_0: iqk_convert_qX_q80_r8<block_q6_0, Q6_0_Dequantizer>(n, vx, bx, vy, nrc_x); break;
case GGML_TYPE_Q8_0: iqk_convert_q80_q80_r8(n, vx, bx, vy, nrc_x); break;
case GGML_TYPE_Q4_0 : iqk_convert_qX_q80_r8<block_q4_0, Q4_0_Dequantizer>(n, vx, bx, vy, nrc_x); break;
case GGML_TYPE_Q5_0 : iqk_convert_qX_q80_r8<block_q5_0, Q5_0_Dequantizer>(n, vx, bx, vy, nrc_x); break;
case GGML_TYPE_Q6_0 : iqk_convert_qX_q80_r8<block_q6_0, Q6_0_Dequantizer>(n, vx, bx, vy, nrc_x); break;
case GGML_TYPE_IQ4_NL: iqk_convert_qX_q80_r8<block_iq4_nl, IQ4_NL0_Dequantizer>(n, vx, bx, vy, nrc_x); break;
case GGML_TYPE_Q8_0 : iqk_convert_q80_q80_r8(n, vx, bx, vy, nrc_x); break;
default: return false;
}
return true;

3
ggml/src/iqk/iqk_mul_mat.cpp
View File

@@ -261,6 +261,7 @@ struct MulMat {
case GGML_TYPE_Q4_0 : return nrc_y >= 32 ? GGML_TYPE_Q8_0_R8 : type;
case GGML_TYPE_Q5_0 : return nrc_y >= 32 ? GGML_TYPE_Q8_0_R8 : type;
case GGML_TYPE_Q6_0 : return nrc_y >= 32 ? GGML_TYPE_Q8_0_R8 : type;
case GGML_TYPE_IQ4_NL : return nrc_y >= 32 ? GGML_TYPE_Q8_0_R8 : type;
case GGML_TYPE_Q8_0 : return nrc_y >= 32 ? GGML_TYPE_Q8_0_R8 : type;
default: break;
}
@@ -413,7 +414,7 @@ bool iqk_convert_repack(int typeA, int n, const void * vx, size_t bx, void * vy,
//case GGML_TYPE_Q5_1:
case GGML_TYPE_Q6_0:
case GGML_TYPE_Q8_0:
//case GGML_TYPE_IQ4_NL:
case GGML_TYPE_IQ4_NL:
//case GGML_TYPE_Q4_0_R8:
//case GGML_TYPE_Q5_0_R4:
//case GGML_TYPE_Q6_0_R4: