mtmd : Support Pixtral 12B (#13065)

* add pixtral text model (vision is wip) * cgraph ok, just missing 2D RoPE * fix bad rebase * first working version * fix problem with img_break token * support dynamic image size * update docs * update test script
2025-04-23 20:21:59 +02:00 · 2025-04-23 20:21:59 +02:00 · ecda2ec4b3
commit ecda2ec4b3
parent eb1776b15a
14 changed files with 643 additions and 31 deletions
--- a/include/llama.h
+++ b/include/llama.h
@ -111,6 +111,7 @@ extern "C" {
        LLAMA_VOCAB_PRE_TYPE_TRILLION       = 31,
        LLAMA_VOCAB_PRE_TYPE_BAILINGMOE     = 32,
        LLAMA_VOCAB_PRE_TYPE_LLAMA4         = 33,
+        LLAMA_VOCAB_PRE_TYPE_PIXTRAL        = 34,
    };

    enum llama_rope_type {