ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-30 19:31:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

Some tweaks for iq2_k and iq3_k

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-09-13 19:50:46 +03:00
parent f77de99df9
commit cb369c22dd
1 changed files with 5 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/llama.cpp
View File

@@ -15619,7 +15619,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
} }
else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS || ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS || else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XXS || ftype == LLAMA_FTYPE_MOSTLY_IQ2_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS ||
ftype == LLAMA_FTYPE_MOSTLY_IQ1_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M || ftype == LLAMA_FTYPE_MOSTLY_IQ1_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M ||
ftype == LLAMA_FTYPE_MOSTLY_IQ1_M || ftype == LLAMA_FTYPE_MOSTLY_IQ2_K) { ftype == LLAMA_FTYPE_MOSTLY_IQ1_M || ftype == LLAMA_FTYPE_MOSTLY_IQ2_K || ftype == LLAMA_FTYPE_MOSTLY_IQ3_K) {
new_type = !qs.has_output ? GGML_TYPE_IQ4_K : GGML_TYPE_Q5_K; new_type = !qs.has_output ? GGML_TYPE_IQ4_K : GGML_TYPE_Q5_K;
} }
else if ((ftype == LLAMA_FTYPE_MOSTLY_IQ3_S || ftype == LLAMA_FTYPE_MOSTLY_IQ3_M || ftype == LLAMA_FTYPE_MOSTLY_IQ4_XS) && !qs.has_output) { else if ((ftype == LLAMA_FTYPE_MOSTLY_IQ3_S || ftype == LLAMA_FTYPE_MOSTLY_IQ3_M || ftype == LLAMA_FTYPE_MOSTLY_IQ4_XS) && !qs.has_output) {
@@ -15640,9 +15640,6 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) { else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) {
new_type = GGML_TYPE_IQ3_S; new_type = GGML_TYPE_IQ3_S;
} }
//else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_K) {
// new_type = GGML_TYPE_IQ3_K;
//}
else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) { else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) {
new_type = GGML_TYPE_IQ3_S; new_type = GGML_TYPE_IQ3_S;
} }
@@ -15687,8 +15684,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
new_type = qs.model.hparams.n_gqa() >= 4 ? GGML_TYPE_Q4_K : GGML_TYPE_Q3_K; new_type = qs.model.hparams.n_gqa() >= 4 ? GGML_TYPE_Q4_K : GGML_TYPE_Q3_K;
} }
else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_K) { else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_K) {
//if (use_more_bits(qs.i_attention_wv, qs.n_attention_wv)) new_type = GGML_TYPE_IQ4_K; new_type = qs.model.hparams.n_gqa() >= 2 ? GGML_TYPE_IQ4_K : GGML_TYPE_IQ3_K;
new_type = GGML_TYPE_IQ3_K;
} }
else if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S && qs.model.hparams.n_gqa() >= 4) { else if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S && qs.model.hparams.n_gqa() >= 4) {
new_type = GGML_TYPE_Q4_K; new_type = GGML_TYPE_Q4_K;
@@ -15700,6 +15696,9 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
else if ((ftype == LLAMA_FTYPE_MOSTLY_IQ3_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ3_S || ftype == LLAMA_FTYPE_MOSTLY_IQ3_K) && qs.model.hparams.n_gqa() >= 2) { else if ((ftype == LLAMA_FTYPE_MOSTLY_IQ3_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ3_S || ftype == LLAMA_FTYPE_MOSTLY_IQ3_K) && qs.model.hparams.n_gqa() >= 2) {
new_type = GGML_TYPE_IQ4_K; new_type = GGML_TYPE_IQ4_K;
} }
else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_K && qs.model.hparams.n_gqa() >= 2) {
new_type = GGML_TYPE_IQ4_K;
}
else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_M) { else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_M) {
new_type = qs.model.hparams.n_gqa() >= 2 ? GGML_TYPE_IQ5_K : GGML_TYPE_IQ4_K; new_type = qs.model.hparams.n_gqa() >= 2 ? GGML_TYPE_IQ5_K : GGML_TYPE_IQ4_K;
} }