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

Adding IQ1_KT - 1.75 bpw SOTA quants (#616)

Browse Source 
* iq1_kt: basics

* iq1_kt: CUDA dequantize

Testing with LlaMA-3.1-8B-Instruct, we get almost the same PPL
as iq2_xxs, so about 0.2 bpw fewer bits for the same quality.

* iq1_kt: CUDA MMQ

* iq1_kt: CUDA MMVQ

* iq1_kt: AVX2 GEMM/GEMV

* iq1_kt: convert/repack to q8_0_r8 (AVX2)

* iq1_kt: slightly faster GEMV

18.6 t/s -> 19.4 t/s

* iq1_kt: NEON GEMM/GEMV

Pathetic as usual

* iq1_kt: slightly faster NEON - still pathetic

* iq1_kt: tiny bit better GEMV on NEON

* iq1_kt: convert/repack to q8_0_r8 (NEON)

* iq1_kt: very slightly faster convert/repack to q8_0_r8 on NEON

* Adding frgotten file

* iq1_kt: add to constants.py

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-07-20 10:05:23 +02:00
committed by GitHub
parent 07673c6c33
commit f989fb03bd
21 changed files with 930 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
src/llama.cpp
View File

@@ -4429,6 +4429,7 @@ struct llama_model_loader {
case GGML_TYPE_IQ2_S_R4:ftype = LLAMA_FTYPE_MOSTLY_IQ2_M_R4;break;
case GGML_TYPE_IQ3_XXS: ftype = LLAMA_FTYPE_MOSTLY_IQ3_XXS; break;
case GGML_TYPE_IQ3_XXS_R4: ftype = LLAMA_FTYPE_MOSTLY_IQ3_XXS_R4; break;
case GGML_TYPE_IQ1_KT: ftype = LLAMA_FTYPE_MOSTLY_IQ1_KT; break;
case GGML_TYPE_IQ2_KT: ftype = LLAMA_FTYPE_MOSTLY_IQ2_KT; break;
case GGML_TYPE_IQ3_KT: ftype = LLAMA_FTYPE_MOSTLY_IQ3_KT; break;
case GGML_TYPE_IQ4_KT: ftype = LLAMA_FTYPE_MOSTLY_IQ4_KT; break;
@@ -5174,6 +5175,7 @@ static std::string llama_model_ftype_name(llama_ftype ftype) {
case LLAMA_FTYPE_MOSTLY_IQ2_M_R4: return "IQ2_M_R4 - 2.7 bpw";
case LLAMA_FTYPE_MOSTLY_IQ3_XS: return "IQ3_XS - 3.3 bpw";
case LLAMA_FTYPE_MOSTLY_IQ3_XXS: return "IQ3_XXS - 3.0625 bpw";
case LLAMA_FTYPE_MOSTLY_IQ1_KT: return "IQ1_KT - 1.75 bpw";
case LLAMA_FTYPE_MOSTLY_IQ2_KT: return "IQ2_KT - 2.125 bpw";
case LLAMA_FTYPE_MOSTLY_IQ3_KT: return "IQ3_KT - 3.125 bpw";
case LLAMA_FTYPE_MOSTLY_IQ4_KT: return "IQ4_KT - 4.0 bpw";
@@ -19170,7 +19172,8 @@ static ggml_type change_type_if_necessary(ggml_type new_type, int nx, int ny) {
new_type == GGML_TYPE_IQ3_XXS_R4 || new_type == GGML_TYPE_IQ2_XXS_R4 || new_type == GGML_TYPE_IQ2_XS_R4 ||
new_type == GGML_TYPE_IQ2_S_R4|| new_type == GGML_TYPE_IQ3_S_R4|| new_type == GGML_TYPE_IQ3_KS ||
new_type == GGML_TYPE_IQ2_KT || new_type == GGML_TYPE_IQ3_KT || new_type == GGML_TYPE_IQ4_KT ||
new_type == GGML_TYPE_IQ5_KS || new_type == GGML_TYPE_IQ5_KS_R4|| new_type == GGML_TYPE_IQ2_KL) {
new_type == GGML_TYPE_IQ5_KS || new_type == GGML_TYPE_IQ5_KS_R4|| new_type == GGML_TYPE_IQ2_KL ||
new_type == GGML_TYPE_IQ1_KT) {
if (nx % QK_K != 0) {
LLAMA_LOG_WARN("\n\n%s : tensor cols %d x %d are not divisible by %d, required for %s", __func__, nx, ny, QK_K, ggml_type_name(new_type));
convert_incompatible_tensor = true;
@@ -19210,6 +19213,7 @@ static ggml_type change_type_if_necessary(ggml_type new_type, int nx, int ny) {
case GGML_TYPE_IQ4_KS:
case GGML_TYPE_IQ4_KS_R4:
case GGML_TYPE_IQ4_XS_R8:
case GGML_TYPE_IQ1_KT:
case GGML_TYPE_IQ2_KT:
case GGML_TYPE_IQ3_KT:
case GGML_TYPE_IQ4_KT:
@@ -19342,7 +19346,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
ftype == LLAMA_FTYPE_MOSTLY_IQ2_K_R4 || ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS_R4 || ftype == LLAMA_FTYPE_MOSTLY_IQ2_KL ||
ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS_R4 || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M_R4 ||
ftype == LLAMA_FTYPE_MOSTLY_IQ1_S_R4 || ftype == LLAMA_FTYPE_MOSTLY_IQ1_M_R4 ||
ftype == LLAMA_FTYPE_MOSTLY_IQ2_KT || ftype == LLAMA_FTYPE_MOSTLY_IQ3_KT) {
ftype == LLAMA_FTYPE_MOSTLY_IQ2_KT || ftype == LLAMA_FTYPE_MOSTLY_IQ3_KT || ftype == LLAMA_FTYPE_MOSTLY_IQ1_KT) {
new_type = !qs.has_output ? GGML_TYPE_IQ4_K : GGML_TYPE_Q5_K;
}
else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS_R4 || ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS_R4) {
@@ -19936,6 +19940,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
case LLAMA_FTYPE_MOSTLY_IQ2_XS: default_type = GGML_TYPE_IQ2_XS; break;
case LLAMA_FTYPE_MOSTLY_IQ2_XS_R4:default_type = GGML_TYPE_IQ2_XS_R4; break;
case LLAMA_FTYPE_MOSTLY_IQ2_KS: default_type = GGML_TYPE_IQ2_KS; break;
case LLAMA_FTYPE_MOSTLY_IQ1_KT: default_type = GGML_TYPE_IQ1_KT; break;
case LLAMA_FTYPE_MOSTLY_IQ2_KT: default_type = GGML_TYPE_IQ2_KT; break;
case LLAMA_FTYPE_MOSTLY_IQ2_S: default_type = GGML_TYPE_IQ2_XS; break;
case LLAMA_FTYPE_MOSTLY_IQ2_M: default_type = GGML_TYPE_IQ2_S; break;