diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
index 88eab7b4dc84..def72aa53dc6 100644
--- a/src/llama-arch.cpp
+++ b/src/llama-arch.cpp
@@ -340,6 +340,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_ROPE_SCALING_ATTN_FACTOR,       "%s.rope.scaling.attn_factor"             },
     { LLM_KV_ROPE_SCALING_ORIG_CTX_LEN,      "%s.rope.scaling.original_context_length" },
     { LLM_KV_ROPE_SCALING_FINETUNED,         "%s.rope.scaling.finetuned"               },
+    { LLM_KV_ROPE_SCALING_YARN_TRUNCATE,     "%s.rope.scaling.yarn_truncate"           },
     { LLM_KV_ROPE_SCALING_YARN_LOG_MUL,      "%s.rope.scaling.yarn_log_multiplier"     },
     { LLM_KV_ROPE_SCALING_YARN_EXT_FACTOR,   "%s.rope.scaling.yarn_ext_factor"         },
     { LLM_KV_ROPE_SCALING_YARN_ATTN_FACTOR,  "%s.rope.scaling.yarn_attn_factor"        },
diff --git a/src/llama-arch.h b/src/llama-arch.h
index 2d8ad6b07702..d7c6a7a0aaf8 100644
--- a/src/llama-arch.h
+++ b/src/llama-arch.h
@@ -345,6 +345,7 @@ enum llm_kv {
     LLM_KV_ROPE_SCALING_ATTN_FACTOR,
     LLM_KV_ROPE_SCALING_ORIG_CTX_LEN,
     LLM_KV_ROPE_SCALING_FINETUNED,
+    LLM_KV_ROPE_SCALING_YARN_TRUNCATE,
     LLM_KV_ROPE_SCALING_YARN_LOG_MUL,
     LLM_KV_ROPE_SCALING_YARN_EXT_FACTOR,
     LLM_KV_ROPE_SCALING_YARN_ATTN_FACTOR,
diff --git a/src/llama-hparams.h b/src/llama-hparams.h
index 064de68f36c2..d9b5d98e8ee4 100644
--- a/src/llama-hparams.h
+++ b/src/llama-hparams.h
@@ -63,6 +63,7 @@ struct llama_hparams {
     bool vocab_only;
     bool no_alloc;
     bool rope_finetuned;
+    bool yarn_truncate = true;
     bool use_par_res;
     bool swin_norm;
     bool norm_before_residual = false;
diff --git a/src/llama-kv-cache.cpp b/src/llama-kv-cache.cpp
index 463510caa768..a3b62b6bc558 100644
--- a/src/llama-kv-cache.cpp
+++ b/src/llama-kv-cache.cpp
@@ -1974,16 +1974,23 @@ ggml_tensor * llama_kv_cache::build_rope_shift(
         tmp = ggml_rope_ext(ctx, tmp,
                 shift, factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                 yarn_ext_factor, yarn_attn_factor, yarn_beta_fast, yarn_beta_slow);
+        ggml_rope_set_truncate(tmp, hparams.yarn_truncate);
 
         // rotate fwd
         tmp = llama_mul_mat_hadamard(ctx, tmp, rot);
 
         tmp = ggml_cpy(ctx, tmp, cur);
-    } else {
+    } else if (hparams.yarn_truncate) {
         // we rotate only the first n_rot dimensions
         tmp = ggml_rope_ext_inplace(ctx, cur,
                 shift, factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                 yarn_ext_factor, yarn_attn_factor, yarn_beta_fast, yarn_beta_slow);
+    } else {
+        tmp = ggml_rope_ext(ctx, cur,
+                shift, factors, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
+                yarn_ext_factor, yarn_attn_factor, yarn_beta_fast, yarn_beta_slow);
+        ggml_rope_set_truncate(tmp, false);
+        tmp = ggml_cpy(ctx, tmp, cur);
     }
 
     return tmp;
diff --git a/src/llama-model-saver.cpp b/src/llama-model-saver.cpp
index 1c7cbd9005d9..c9ae3319f9c4 100644
--- a/src/llama-model-saver.cpp
+++ b/src/llama-model-saver.cpp
@@ -374,6 +374,7 @@ void llama_model_saver::add_kv_from_model() {
     add_kv(LLM_KV_ROPE_SCALING_ATTN_FACTOR,          hparams.rope_attn_factor);
     add_kv(LLM_KV_ROPE_SCALING_ORIG_CTX_LEN,         hparams.n_ctx_orig_yarn);
     add_kv(LLM_KV_ROPE_SCALING_FINETUNED,            hparams.rope_finetuned);
+    add_kv(LLM_KV_ROPE_SCALING_YARN_TRUNCATE,        hparams.yarn_truncate);
     add_kv(LLM_KV_ROPE_SCALING_YARN_LOG_MUL,         hparams.rope_yarn_log_mul);
     add_kv(LLM_KV_ROPE_SCALING_YARN_EXT_FACTOR,      hparams.yarn_ext_factor);
     add_kv(LLM_KV_ROPE_SCALING_YARN_ATTN_FACTOR,     hparams.yarn_attn_factor);
diff --git a/src/models/plamo3.cpp b/src/models/plamo3.cpp
index f8235f8a8323..c818742b1e3e 100644
--- a/src/models/plamo3.cpp
+++ b/src/models/plamo3.cpp
@@ -2,6 +2,7 @@
 
 void llama_model_plamo3::load_arch_hparams(llama_model_loader & ml) {
     ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
+    ml.get_key(LLM_KV_ROPE_SCALING_YARN_TRUNCATE, hparams.yarn_truncate, false);
     const bool found_swa = ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa, false);
     if (found_swa && hparams.n_swa > 0) {
         hparams.swa_type = LLAMA_SWA_TYPE_STANDARD;
@@ -131,9 +132,11 @@ llama_model_plamo3::graph<iswa>::graph(const llama_model & model, const llm_grap
         Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr,
                 n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
                 ext_factor, attn_factor, beta_fast, beta_slow);
+        ggml_rope_set_truncate(Qcur, hparams.yarn_truncate);
         Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr,
                 n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
                 ext_factor, attn_factor, beta_fast, beta_slow);
+        ggml_rope_set_truncate(Kcur, hparams.yarn_truncate);
 
         const float attn_scale = 1.0f / sqrtf(float(head_dim_q));