root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-11-15 15:29:53 +00:00
Code Issues Actions 13 Packages Projects Releases Wiki Activity

convert_hf : prefer SentencePiece tokenizer for Mamba-2 when present

Browse Source 
The tokenzier.json of Mamba-Codestral-7B-v0.1 otherwise requires
workarounds to work correctly.
This commit is contained in:
Francis Couture-Harpin 2024-08-22 14:31:12 -04:00
parent fa358e7071
commit 38913dc8dd
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
convert_hf_to_gguf.py
View File

@ -2801,13 +2801,13 @@ class Mamba2Model(Model):
vocab_size = -(vocab_size // -pad_vocab) * pad_vocab vocab_size = -(vocab_size // -pad_vocab) * pad_vocab
self.hparams["vocab_size"] = vocab_size self.hparams["vocab_size"] = vocab_size
if (self.dir_model / "tokenizer.json").is_file(): if (self.dir_model / "tokenizer.model").is_file():
self._set_vocab_gpt2()
elif (self.dir_model / "tokenizer.model").is_file():
self._set_vocab_sentencepiece() self._set_vocab_sentencepiece()
elif (self.dir_model / "tokenizer.model.v3").is_file(): elif (self.dir_model / "tokenizer.model.v3").is_file():
# mamba-codestral # mamba-codestral
raise NotImplementedError(f"Please rename {self.dir_model / 'tokenizer.model.v3'} to {self.dir_model / 'tokenizer.model'}") 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: else:
# Use the GPT-NeoX tokenizer when no tokenizer files are present # Use the GPT-NeoX tokenizer when no tokenizer files are present
self._set_vocab_builtin("gpt-neox", vocab_size) self._set_vocab_builtin("gpt-neox", vocab_size)