happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

refactor: remove redundant deepseek2 compatibility code

This commit is contained in:
suhyun-hwang 2026-01-13 23:22:34 +09:00
parent d85a08830b
commit ab233049dc
1 changed files with 8 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
src/llama-model.cpp
View File

@ -1134,14 +1134,11 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_ATTENTION_Q_LORA_RANK, hparams.n_lora_q);
ml.get_key(LLM_KV_ATTENTION_KV_LORA_RANK, hparams.n_lora_kv);
ml.get_key(LLM_KV_ATTENTION_KEY_LENGTH_MLA, hparams.n_embd_head_k_mla, false);
ml.get_key(LLM_KV_ATTENTION_VALUE_LENGTH_MLA, hparams.n_embd_head_v_mla, false);
if (hparams.n_embd_head_k_mla != 0 && hparams.n_embd_head_v_mla != 0) {
hparams.n_embd_head_k = hparams.n_embd_head_k_mla;
hparams.n_embd_head_v = hparams.n_embd_head_v_mla;
hparams.n_head_kv_arr = hparams.n_head_arr;
}
ml.get_key(LLM_KV_ATTENTION_KEY_LENGTH_MLA, hparams.n_embd_head_k_mla);
ml.get_key(LLM_KV_ATTENTION_VALUE_LENGTH_MLA, hparams.n_embd_head_v_mla);
hparams.n_embd_head_k = hparams.n_embd_head_k_mla;
hparams.n_embd_head_v = hparams.n_embd_head_v_mla;
hparams.n_head_kv_arr = hparams.n_head_arr;
ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT, hparams.n_layer_dense_lead);
ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp);
@ -1155,10 +1152,9 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa, false);
ml.get_key(LLM_KV_ROPE_FREQ_BASE_SWA, hparams.rope_freq_base_train_swa, false);
if (hparams.n_swa > 0) {
hparams.swa_type = LLAMA_SWA_TYPE_STANDARD;
ml.get_key_or_arr(LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN, hparams.swa_layers, hparams.n_layer);
}
hparams.swa_type = LLAMA_SWA_TYPE_STANDARD;
ml.get_key_or_arr(LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN, hparams.swa_layers, hparams.n_layer);
switch (hparams.n_layer) {
case 24: type = LLM_TYPE_7B_A1B; break;