upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-10-03 11:27:25 -05:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'onyx' into onyx-vision

Browse Source
This commit is contained in:
Pedro Cuenca
2026-08-08 09:40:03 +02:00
parent 1f2a51f268 fe54c4f6f6
commit 4037e13ea9
1 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/onyx.cpp
+5
View File
@@ -7,6 +7,11 @@ void llama_model_onyx::load_arch_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_LOGIT_SCALE, hparams.f_logit_scale);
ml.get_key(LLM_KV_ATTENTION_POST_NORM_RMS_EPS, hparams.f_post_norm_rms_eps);
// SWA layers share the model rope theta; they are also the only layers that use rope
// here (global layers are NoPE), so the 10000.0 default would apply to all of them.
hparams.rope_freq_base_train_swa = hparams.rope_freq_base_train;
ml.get_key(LLM_KV_ROPE_FREQ_BASE_SWA, hparams.rope_freq_base_train_swa, false);
// SWA + NoPE: [SW, SW, SW, Full], NoPE used on Full layers.
if (hparams.n_swa > 0) {
hparams.swa_type = LLAMA_SWA_TYPE_STANDARD;