diff options
| author | Kawrakow <iwankawrakow@gmail.com> | 2025-07-20 10:05:23 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-07-20 10:05:23 +0200 |
| commit | f989fb03bd12752ad6e93717ca4bd298d5001d99 (patch) | |
| tree | 7a127aba5c05667904b7e28a46d07c2d295ef619 /src/llama.cpp | |
| parent | 07673c6c33753487dd054dcff37f19d93d6c56d3 (diff) | |
Adding IQ1_KT - 1.75 bpw SOTA quants (#616)
* 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>
Diffstat (limited to 'src/llama.cpp')
| -rw-r--r-- | src/llama.cpp | 9 |
1 files changed, 7 insertions, 2 deletions
diff --git a/src/llama.cpp b/src/llama.cpp index 0d29f24a..2c251c6b 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -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; |