feat: Add late interaction model training support for retrieval#2283
Draft
rnyak wants to merge 5 commits into
Draft
feat: Add late interaction model training support for retrieval#2283rnyak wants to merge 5 commits into
rnyak wants to merge 5 commits into
Conversation
rnyak
requested review from
a team,
HuiyingLi,
adil-a,
akoumpa and
hemildesai
as code owners
Collaborator
Author
|
/claude review |
Signed-off-by: Ronay Ak <ronaya@nvidia.com>
Signed-off-by: Ronay Ak <ronaya@nvidia.com>
…egatives Signed-off-by: Ronay Ak <ronaya@nvidia.com>
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What does this PR do ?
Updated
train_bi_encoder.pyto support local ColBERT-style (multi_vector) pooling by addingcolbert_scores_and_labels(), which computes MaxSim scores with query and passage attention-mask handling. The train and validation paths now route ColBERT models through this scoring function instead of standard pooled embedding contrastive scoring.The changes in the MR add support for
multi_vectorpooling and maxsim scoring with and without distributed in-batch neg training.The latest commits:
Added
detach_distributed_inbatch_negativesarg (default True), so distributed in-batch negatives can use the previous efficient detached behavior by default, or preserve remote passage gradients when set to false.Renamed ColBERT-style scoring to multi_vector/MaxSim, kept colbert as a backward-compatible alias, and made distributed MaxSim scoring more memory efficient by scoring one passage slot at a time.
Before your PR is "Ready for review"
Pre checks: