@@ -16275,8 +16275,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
1627516275 if (ftype == LLAMA_FTYPE_MOSTLY_Q3_K_L) new_type = GGML_TYPE_Q4_K;
1627616276 }
1627716277 ++qs.i_attention_wo;
16278- }
16279- else if (name.find("attn_qkv.weight") != std::string::npos) {
16278+ } else if (name.find("attn_qkv.weight") != std::string::npos) {
1628016279 if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S || ftype == LLAMA_FTYPE_MOSTLY_Q2_K || ftype == LLAMA_FTYPE_MOSTLY_Q2_K_L) {
1628116280 new_type = GGML_TYPE_Q3_K;
1628216281 }
@@ -16300,8 +16299,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
1630016299 else if (ftype == LLAMA_FTYPE_MOSTLY_IQ4_XS) new_type = GGML_TYPE_Q4_K;
1630116300 else if (ftype == LLAMA_FTYPE_MOSTLY_IQ4_XSR || ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXXL) new_type = GGML_TYPE_Q5_K;
1630216301 ++qs.i_attention_wv;
16303- }
16304- else if (name.find("ffn_gate") != std::string::npos) {
16302+ } else if (name.find("ffn_gate") != std::string::npos) {
1630516303 auto info = layer_info(qs.i_ffn_gate, qs.n_ffn_gate, name.c_str());
1630616304 int i_layer = info.first, n_layer = info.second;
1630716305 if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K_L && (difquant_half_tensors(i_layer, n_layer))) new_type = GGML_TYPE_Q3_K;
@@ -16319,8 +16317,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
1631916317 else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXL && (difquant_half_tensors(i_layer, n_layer))) new_type = GGML_TYPE_IQ4_XS;
1632016318 else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXXL && (difquant_six_eights_tensors(i_layer, n_layer))) new_type = GGML_TYPE_IQ4_XS;
1632116319 ++qs.i_ffn_gate;
16322- }
16323- else if (name.find("ffn_up") != std::string::npos) {
16320+ } else if (name.find("ffn_up") != std::string::npos) {
1632416321 auto info = layer_info(qs.i_ffn_up, qs.n_ffn_up, name.c_str());
1632516322 int i_layer = info.first, n_layer = info.second;
1632616323 if (ftype == LLAMA_FTYPE_MOSTLY_Q2_K_L && (difquant_half_tensors(i_layer, n_layer))) new_type = GGML_TYPE_Q3_K;
0 commit comments