Commit 7ceed8737 for llama.cpp

commit 7ceed8737fdb4eb09b4760e77bd12d38012de5a8
Author: Kartik Gulia <kgulia@nvidia.com>
Date:   Wed Sep 16 18:54:47 2026 +0530

    models : allow Nemotron-H models to only define layer_norm_epsilon (#28989)

    * allows nemotron models to get by with just defining layer_norm_epsilon

    * made changes to load_arch_hparams instead

diff --git a/src/models/nemotron-h.cpp b/src/models/nemotron-h.cpp
index ff8784d18..24ed9a673 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 layer
     ml.get_key_or_arr(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp_arr, hparams.n_layer_all, false);