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

Fix very low bpw missing imatrix check (#1284)

Browse Source
This commit is contained in:
Kawrakow
2026-02-19 08:15:26 +01:00
committed by GitHub
parent 51df09be8a
commit d81cde5cea
1 changed files with 21 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
src/llama-quantize.cpp
View File

@@ -1425,23 +1425,27 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
}
}
}
if (!params->ignore_imatrix_rules && !imatrix &&
(new_type == GGML_TYPE_IQ2_XXS ||
new_type == GGML_TYPE_IQ2_XXS_R4 ||
new_type == GGML_TYPE_IQ2_XS ||
new_type == GGML_TYPE_IQ2_XS_R4 ||
new_type == GGML_TYPE_IQ2_S ||
new_type == GGML_TYPE_IQ2_S_R4||
new_type == GGML_TYPE_IQ1_S ||
new_type == GGML_TYPE_IQ1_S_R4||
new_type == GGML_TYPE_IQ1_M_R4||
(new_type == GGML_TYPE_IQ1_M && strcmp(tensor->name, "token_embd.weight") && strcmp(tensor->name, "output.weight")) ||
(new_type == GGML_TYPE_Q2_K && ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S && strcmp(tensor->name, "token_embd.weight") != 0))) {
LLAMA_LOG_ERROR("\n\n============================================================\n");
LLAMA_LOG_ERROR("Missing importance matrix for tensor %s in a very low-bit quantization\n", tensor->name);
LLAMA_LOG_ERROR("The result will be garbage, so bailing out\n");
LLAMA_LOG_ERROR("============================================================\n\n");
throw std::runtime_error(format("Missing importance matrix for tensor %s in a very low-bit quantization", tensor->name));
if (!params->ignore_imatrix_rules && !imatrix) {
bool is_very_low_bpw_quant = new_type == GGML_TYPE_IQ2_XXS ||
new_type == GGML_TYPE_IQ2_XXS_R4 ||
new_type == GGML_TYPE_IQ2_XS ||
new_type == GGML_TYPE_IQ2_XS_R4 ||
new_type == GGML_TYPE_IQ2_S ||
new_type == GGML_TYPE_IQ2_S_R4 ||
new_type == GGML_TYPE_IQ1_S ||
new_type == GGML_TYPE_IQ1_S_R4 ||
new_type == GGML_TYPE_IQ1_M ||
new_type == GGML_TYPE_IQ1_M_R4 ||
new_type == GGML_TYPE_IQ1_KT ||
new_type == GGML_TYPE_IQ2_KT ||
(new_type == GGML_TYPE_Q2_K && ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S);
if (is_very_low_bpw_quant && strcmp(tensor->name, "token_embd.weight") && strcmp(tensor->name, "output.weight")) {
LLAMA_LOG_ERROR("\n\n============================================================\n");
LLAMA_LOG_ERROR("Missing importance matrix for tensor %s in a very low-bit quantization\n", tensor->name);
LLAMA_LOG_ERROR("The result will be garbage, so bailing out\n");
LLAMA_LOG_ERROR("============================================================\n\n");
throw std::runtime_error(format("Missing importance matrix for tensor %s in a very low-bit quantization", tensor->name));
}
}
float * f32_data;