add eos handling #9
Draft
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
chanr0
commented
Jun 6, 2025
|
|
||
| # construct mappings from byte vocab to indices in weight array | ||
| self.trie_decode = ( | ||
| list(range(256)) + atomic_tokens + self.eos_tokens + [self.eot_token] |
chanr0
commented
Jun 6, 2025
| ) | ||
| self.trie_encode = { | ||
| k: v for k, v in zip(self.trie_decode, list(range(len(self.trie_decode)))) | ||
| } |
Author
There was a problem hiding this comment.
add a check for encode that there are no duplicates! (that decode doesn't have duplicate values)
chanr0
commented
Jun 6, 2025
| return self.materialize().map_keys( | ||
| lambda x: bytes([x]) if x is not None else "EOT" | ||
| lambda x: bytes([x]) if x in range(256) else ("EOT" if x is None else x) | ||
| ) |
Author
There was a problem hiding this comment.
handle EOT being something other than None, e.g., decode[-1]
chanr0
commented
Jun 6, 2025
| return y.replace(" ", "␣") | ||
|
|
||
|
|
||
| def split_with_atomic_tokens(data: bytes, atomic_tokens: list[bytes]) -> list[Union[int, bytes]]: |
Author
There was a problem hiding this comment.
@benlebrun need to check with tim what the best handling is
chanr0
commented
Jun 6, 2025
| """ | ||
| if node := self.children[self.node].get(b): | ||
| mass = self.mass | ||
| if b in self.trie.trie.eos_tokens: |
Author
There was a problem hiding this comment.
not good, handle during trie construction. For instance, assume your eos tokens are 'ab', 'cd'
- during prefill, you want to have the a-b, c-d, i.e., byte-level nodes
- during sampling, you want to group them both into a single EOS node
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
add eos handling