feat: Add granite-docling preprocessor workaround mapping

It uses the same pretokenizer as granite3, but the vocab is different so we
need a different chkhsh.

Branch: GraniteDocling

Signed-off-by: Gabe Goodhart <[email protected]>

Author: gabe-l-hart

convert_hf_to_gguf.py

@@ -738,6 +738,9 @@ def get_vocab_base_pre(self, tokenizer) -> str:
         if chkhsh == "66b8d4e19ab16c3bfd89bce5d785fb7e0155e8648708a1f42077cb9fe002c273":
             # ref: https://huggingface.co/alvarobartt/grok-2-tokenizer
             res = "grok-2"
+        if chkhsh == "53e325976a6e142379c19b09afcae354f2f496f147afa8f9e189a33fe4e3024e":
+            # ref: https://huggingface.co/ibm-granite/granite-docling-258M
+            res = "refact"
         if chkhsh == "0ef9807a4087ebef797fc749390439009c3b9eda9ad1a097abbe738f486c01e5":
             # ref: https://huggingface.co/meta-llama/Meta-Llama-3-8B
             res = "llama-bpe"

convert_hf_to_gguf_update.py

@@ -160,6 +160,8 @@ class TOKENIZER_TYPE(IntEnum):
     {"name": "kimi-k2",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/moonshotai/Kimi-K2-Base",   "chkhsh": "81212dc7cdb7e0c1074ca62c5aeab0d43c9f52b8a737be7b12a777c953027890"},
     {"name": "qwen2",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen3-Embedding-0.6B", "chkhsh": "d4540891389ea895b53b399da6ac824becc30f2fba0e9ddbb98f92e55ca0e97c"},
     {"name": "grok-2",    "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/alvarobartt/grok-2-tokenizer", "chkhsh": "66b8d4e19ab16c3bfd89bce5d785fb7e0155e8648708a1f42077cb9fe002c273"},
+    # granite-docling uses granite3 tokenizer w/ additional tokens => refact pretokenizer
+    {"name": "refact",    "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ibm-granite/granite-docling-258M", "chkhsh": "53e325976a6e142379c19b09afcae354f2f496f147afa8f9e189a33fe4e3024e"},
 ]