LLM inference in C/C++
C++ ggml
models : allow Nemotron-H models to only define layer_norm_epsilon (#28989)
Kartik Gulia7ceed8737fdb
diff --git a/src/models/nemotron-h.cpp b/src/models/nemotron-h.cppindex ff8784d182..24ed9a673e 100644--- a/src/models/nemotron-h.cpp+++ b/src/models/nemotron-h.cpp@@ -15,8 +15,10 @@ void llama_model_nemotron_h::load_arch_hparams(llama_model_loader & ml) {hparams.is_recr_impl[i] = i < hparams.n_layer() && hparams.n_head_kv(i) == 0 && hparams.n_ff(i) == 0;}- ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);- ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); // MTP head final_layernorm+ ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); // MTP head final_layernorm+ if (!ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps, false)) {+ hparams.f_norm_rms_eps = hparams.f_norm_eps;+ }// Puzzle models set a different expert FFN size per layerml.get_key_or_arr(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp_arr, hparams.n_layer_all, false);