Commit 72f4603c7 for llama.cpp

commit 72f4603c7c46b43fcec6d6761aeb488c4f382a64
Author: Pascal <admin@serveurperso.com>
Date:   Thu Nov 12 12:14:50 2026 +0100

    mtmd: read the MiniCPM-V projection width from ne[1] (#32132)

    The MiniCPM-V resampler ends with build_mm(mm_model_proj), whose output
    width is ne[1], while n_mmproj_embd() returns ne[0]. Both are equal in
    real models, the projection is square. A non square projection passes
    the text model n_embd check and only aborts at the encode, on the
    output buffer size check.

    Return ne[1], so the width matches the graph output.

diff --git a/tools/mtmd/mtmd-model.cpp b/tools/mtmd/mtmd-model.cpp
index 6cd9509f9..b6727a493 100644
--- a/tools/mtmd/mtmd-model.cpp
+++ b/tools/mtmd/mtmd-model.cpp
@@ -4233,7 +4233,7 @@ int mtmd_model::n_mmproj_embd() const {
         case PROJECTOR_TYPE_MLP_NORM:
             return mm_3_b->ne[0];
         case PROJECTOR_TYPE_MINICPMV:
-            return mm_model_proj->ne[0];
+            return mm_model_proj->ne[1];
         case PROJECTOR_TYPE_MINICPMV4_6:
         case PROJECTOR_TYPE_MINICPMV4_7:
         case PROJECTOR_TYPE_MINICPM_OCR_1_0: