From baf3cc6e1d70ce73f66a0665811e5c2228cddc5d Mon Sep 17 00:00:00 2001 From: Daniel Bevenius Date: Tue, 19 May 2026 18:41:44 +0200 Subject: [PATCH] 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. --- src/models/qwen35.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) 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];