Colbert #248
Merged
Colbert #248
Conversation
This file contains 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
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Outdated suggestions cannot be applied.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.
This PR adds classes for late interaction embeddings.
The only supported model at the moment is colbertv2.0
A few complexities in this PR are bound to the way colbert processes documents and queries.
It adds a special token
[Q]or[D]as a second token in a sequence of tokens.They should be added as token ids to be processed correctly.
Another complexity is that colbert authors recommend to pad queries with [MASK] tokens for query augmentation which leads to a better retrieval quality. Padding extends the query to 32 tokens.
Official colbert library allocates 32 tokens for query + (512 - 32 - 1) for context.
link to query tokenizer
We don't do such a separation, instead we just limit query to 512 tokens.
Colbert also requires removing punctuation tokens, which is being done after inference stage via modifying attention mask and the consequent multiplication of the model's output and updated attention mask.