From: Jonas Jankaitis <redacted>
Date: Mon, 27 Jul 2026 13:21:37 +0000 (+0300)
Subject: fit : count nextn (MTP) blocks in n_gpu_layers so front layers stay on GPU (#26177)
X-Git-Tag: upstream/0.0.10438~286
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=0324696b8e5fe340dc94b64714e4c9aab03084a2;p=pkg%2Fggml%2Fsources%2Fllama.cpp

fit : count nextn (MTP) blocks in n_gpu_layers so front layers stay on GPU (#26177)
---

diff --git a/common/fit.cpp b/common/fit.cpp
index c79221cb0..c82d066ad 100644
--- a/common/fit.cpp
+++ b/common/fit.cpp
@@ -136,7 +136,7 @@ static std::vector<llama_device_memory_data> common_get_device_memory_data_impl(
         devs.push_back(llama_model_get_device(model, i));
     }
 
-    hp_ngl         = llama_model_n_layer(model);
+    hp_ngl         = llama_model_n_layer(model) + llama_model_n_layer_nextn(model);
     hp_n_ctx_train = llama_model_n_ctx_train(model);
     hp_n_expert    = llama_model_n_expert(model);