TokenizerChanger 0.0.3

Library for manipulating the existing tokenizer.

Tokens-Deletion

Python script for manipulating the existing tokenizer.

The solution was tested on Llama3-8B tokenizer.

---

Installation:

Installation from PyPI:

pip install tokenizerchanger

Usage:

changer = TokenizerChanger(tokenizer)

Create the object of TokenizerChanger class that requires an existing tokenizer that will be changed, e.g. PreTrainedTokenizerFast class from рџ¤— tokenizers library.

Deletion:

changer.delete_k_least_frequent_tokens(k=1000)
changer.delete_k_least_frequent_tokens(k=1000, exclude=list_of_tokens)

Deletes k most frequent tokens. The exclude argument stands for tokens that will be ignored during the deletion of least frequent tokens.

changer.delete_unwanted_tokens(list_of_unwanted_tokens)

Deletes all tokens from list_of_unwanted_tokens from the tokenizer.

changer.delete_tokens(list_of_unwanted_tokens)

Now, you can delete exactly the list of unwanted tokens, in contrast to the delete_unwanted_tokens function, which deletes all tokens from the list and tokens that contain unwanted tokens as a substring.

changer.delete_overlaps(vocab)

Finds and deletes all intersections of the tokenizer's vocabulary and the vocab variable from the tokenizer. Notice that vocab should be a dict variable.

changer.delete_inappropriate_merges(vocab)

Deletes all merges from tokenizer which contradict the vocab variable. Notice that vocab should be a list[str] variable.

Addition:

The idea of creating such functions arose due to the fact that the built-in functions do not add tokens/merges properly, when some tokens are deleted. That is why you can get more tokens after encoding the same text, even if the necessary tokens have been added.

changer.add_tokens(list_of_tokens)

Adds the tokens from the list. The indexes will be filled automatically.

changer.add_merges(list_of_merges)

Adds the merges from the list.

"Get" functions:

changer.get_overlapping_tokens(vocab)

Returns the intersection between the tokenizer's vocabulary and the vocab variable. Notice that vocab should be a dict variable.

changer.get_overlapping_megres(merges)

Returns the intersection between the tokenizer's merges and the merges variable. Notice that merges should be a list variable.

Saving:

changer.save_tokenizer(path)

Saves the current state of the changed tokenizer. Additionally, it saves tokenizer configs into path folder (./updated_tokenizer by default).

tokenizer = ch.updated_tokenizer()

Return the changed tokenizer.