vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Simplify via lambda

Browse source
This commit is contained in:
Iwan Kawrakow 2023-06-23 17:50:53 +03:00
parent 333ffcc5ba
commit 88412a1aa0
1 changed files with 6 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

10
llama.cpp
View file

@ -2474,6 +2474,10 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
std::vector<std::thread> workers;
std::mutex mutex;
auto use_more_bits = [] (int i_layer, int num_layers) -> bool {
return i_layer < num_layers/8 || i_layer >= 7*num_layers/8 || (i_layer - num_layers/8)%3 == 2;
};
size_t idx = 0;
for (llama_load_tensor & tensor : model_loader->tensors_map.tensors) {
llama_buffer read_data;
@ -2528,15 +2532,13 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_M || ftype == LLAMA_FTYPE_MOSTLY_Q2_K) new_type = GGML_TYPE_Q4_K;
else if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_L) new_type = GGML_TYPE_Q5_K;
else if ((ftype == LLAMA_FTYPE_MOSTLY_Q4_K_M || ftype == LLAMA_FTYPE_MOSTLY_Q5_K_M) &&
(i_attention_wv < n_attention_wv/8 || i_attention_wv >= 7*n_attention_wv/8 ||
(i_attention_wv - n_attention_wv/8)%3 == 2)) new_type = GGML_TYPE_Q6_K;
use_more_bits(i_attention_wv, n_attention_wv)) new_type = GGML_TYPE_Q6_K;
++i_attention_wv;
} else if (tensor.name.find("feed_forward.w2.weight") != std::string::npos) {
if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_M || ftype == LLAMA_FTYPE_MOSTLY_Q2_K) new_type = GGML_TYPE_Q4_K;
else if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_L) new_type = GGML_TYPE_Q5_K;
else if ((ftype == LLAMA_FTYPE_MOSTLY_Q4_K_M || ftype == LLAMA_FTYPE_MOSTLY_Q5_K_M) &&
(i_feed_forward_w2 < n_feed_forward_w2/8 || i_feed_forward_w2 >= 7*n_feed_forward_w2/8 ||
(i_feed_forward_w2 - n_feed_forward_w2/8)%3 == 2)) new_type = GGML_TYPE_Q6_K;
use_more_bits(i_feed_forward_w2, n_feed_forward_w2)) new_type = GGML_TYPE_Q6_K;
++i_feed_forward_w2;
} else if (tensor.name.find("attention.wo.weight") != std::string::npos) {
if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_M || ftype == LLAMA_FTYPE_MOSTLY_Q2_K) new_type = GGML_TYPE_Q4_K;