From 576c82eda210ca0111c04f5256bf77897a4d4cc4 Mon Sep 17 00:00:00 2001
From: Dowon <ks2515@naver.com>
Date: Fri, 11 Jul 2025 16:36:04 +0900
Subject: [PATCH] vocab : add midm-2.0 model pre-tokenizer (#14626)

---
 convert_hf_to_gguf.py        | 3 +++
 convert_hf_to_gguf_update.py | 1 +
 src/llama-vocab.cpp          | 3 ++-
 3 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
index 52aa87d6a..3d5e7e5a4 100755
--- a/convert_hf_to_gguf.py
+++ b/convert_hf_to_gguf.py
@@ -833,6 +833,9 @@ class TextModel(ModelBase):
         if chkhsh == "48f8e02c0359c0bbdd82f26909171fac1c18a457bb47573ed1fe3bbb2c1cfd4b":
             # ref: https://huggingface.co/tiiuae/Falcon-H1-34B-Base
             res = "falcon-h1"
+        if chkhsh == "f6791d196f87ce6b56a7d234be618e0d58f8cda3549416635b2bebcd22cd95c4":
+            # ref: https://huggingface.co/K-intelligence/Midm-2.0-Base-Instruct
+            res = "midm-2.0"
 
         if res is None:
             logger.warning("\n")
diff --git a/convert_hf_to_gguf_update.py b/convert_hf_to_gguf_update.py
index b8cb6027d..9f9b88da8 100755
--- a/convert_hf_to_gguf_update.py
+++ b/convert_hf_to_gguf_update.py
@@ -129,6 +129,7 @@ models = [
     {"name": "pixtral",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mistral-community/pixtral-12b", },
     {"name": "seed-coder",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ByteDance-Seed/Seed-Coder-8B-Base", },
     {"name": "a.x-4.0",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/skt/A.X-4.0", },
+    {"name": "midm-2.0",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/K-intelligence/Midm-2.0-Base-Instruct", },
 ]
 
 # some models are known to be broken upstream, so we will skip them as exceptions
diff --git a/src/llama-vocab.cpp b/src/llama-vocab.cpp
index 10823b183..02cdc244a 100644
--- a/src/llama-vocab.cpp
+++ b/src/llama-vocab.cpp
@@ -1524,7 +1524,8 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                     tokenizer_pre == "llama-bpe"||
                     tokenizer_pre == "falcon3"  ||
                     tokenizer_pre == "falcon-h1" ||
-                    tokenizer_pre == "pixtral") {
+                    tokenizer_pre == "pixtral"  ||
+                    tokenizer_pre == "midm-2.0") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_LLAMA3;
                 ignore_merges = true;
                 add_bos = true;