From: Daniel Bevenius <redacted>
Date: Tue, 19 May 2026 16:41:44 +0000 (+0200)
Subject: model : clarify MTP layer comment in qwen35.cpp [no ci] (#23338)
X-Git-Tag: upstream/0.0.10438~1202
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=baf3cc6e1d70ce73f66a0665811e5c2228cddc5d;p=pkg%2Fggml%2Fsources%2Fllama.cpp

model : clarify MTP layer comment in qwen35.cpp [no ci] (#23338)

This commit attempts to clarify a code comment in graph_mtp regarding
where the MTP layer is stored.

The motivation for this is that it was not obvious to me what the
original comment meant and hopefully this makes it clearer.
---

diff --git a/src/models/qwen35.cpp b/src/models/qwen35.cpp
index 361d7538a..35a0158e8 100644
--- a/src/models/qwen35.cpp
+++ b/src/models/qwen35.cpp
@@ -496,7 +496,8 @@ llama_model_qwen35::graph_mtp::graph_mtp(const llama_model & model, const llm_gr
     const int64_t n_embd_head = hparams.n_embd_head_v();
     GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());
 
-    // The MTP block lives at the source file's original layer index.
+    // hparams.n_layer includes both main model layers and MTP layers. The MTP
+    // layer is stored immediately after the main layers in model.layers[].
     const int il = (int) hparams.n_layer - (int) hparams.nextn_predict_layers;
     const auto & layer = model.layers[il];