From 38913dc8ddd1e119df0e0cfcacfb260b9b1f5c02 Mon Sep 17 00:00:00 2001 From: Francis Couture-Harpin Date: Thu, 22 Aug 2024 14:31:12 -0400 Subject: [PATCH] convert_hf : prefer SentencePiece tokenizer for Mamba-2 when present The tokenzier.json of Mamba-Codestral-7B-v0.1 otherwise requires workarounds to work correctly. --- convert_hf_to_gguf.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py index a5bdd5def..4851926b7 100755 --- a/convert_hf_to_gguf.py +++ b/convert_hf_to_gguf.py @@ -2801,13 +2801,13 @@ class Mamba2Model(Model): vocab_size = -(vocab_size // -pad_vocab) * pad_vocab self.hparams["vocab_size"] = vocab_size - if (self.dir_model / "tokenizer.json").is_file(): - self._set_vocab_gpt2() - elif (self.dir_model / "tokenizer.model").is_file(): + if (self.dir_model / "tokenizer.model").is_file(): self._set_vocab_sentencepiece() elif (self.dir_model / "tokenizer.model.v3").is_file(): # mamba-codestral raise NotImplementedError(f"Please rename {self.dir_model / 'tokenizer.model.v3'} to {self.dir_model / 'tokenizer.model'}") + elif (self.dir_model / "tokenizer.json").is_file(): + self._set_vocab_gpt2() else: # Use the GPT-NeoX tokenizer when no tokenizer files are present self._set_vocab_builtin("gpt-neox", vocab_size)