From ac670c698540488c818a17c3749abf3ed4cb764a Mon Sep 17 00:00:00 2001 From: Jonas Knobloch Date: Tue, 24 Mar 2026 13:02:49 +0100 Subject: Add tokenizer module * Wrap MBPE tokenizer --- tokenizer/bpe/utility.go | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+) create mode 100644 tokenizer/bpe/utility.go (limited to 'tokenizer/bpe/utility.go') diff --git a/tokenizer/bpe/utility.go b/tokenizer/bpe/utility.go new file mode 100644 index 0000000..8ad0ca2 --- /dev/null +++ b/tokenizer/bpe/utility.go @@ -0,0 +1,20 @@ +package bpe + +import "github.com/jonasknobloch/mbpe" + +func NewTokenizerFromFiles(vocab, merges string) (*Tokenizer, error) { + model := mbpe.NewMBPE() + + if err := model.Load(vocab, merges); err != nil { + return nil, err + } + + tokenizer := mbpe.NewTokenizer(model) + + pre := mbpe.NewByteLevel(false) + + tokenizer.SetPreTokenizer(pre) + tokenizer.SetDecoder(pre) + + return NewTokenizer(tokenizer), nil +} -- cgit v1.3.1