vocab : add midm-2.0 model pre-tokenizer (ggml-org#14626)

Author: Bing-su

convert_hf_to_gguf.py

@@ -833,6 +833,9 @@ def get_vocab_base_pre(self, tokenizer) -> str:
         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")

convert_hf_to_gguf_update.py

@@ -129,6 +129,7 @@ class TOKENIZER_TYPE(IntEnum):
     {"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

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;