BWhat changed
Removed the causal attention mask during contrastive embedding training, used token queries attending a trainable latent key/value dictionary before mean pooling, and staged retrieval then mixed-task instruction tuning.
Scientific scope
Removed the causal attention mask during contrastive embedding training, used token queries attending a trainable latent key/value dictionary before mean pooling, and staged retrieval then mixed-task instruction tuning. Latent dictionary keys/values are not the queries: this is not simply Perceiver latent-query pooling. Second-stage mixed-task training omits in-batch negatives to avoid class false negatives. Public embedding-training data does not mean the pretrained LLM was trained from scratch with that data; snapshot leaderboard results are not current rankings.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Latent dictionary keys/values are not the queries: this is not simply Perceiver latent-query pooling. Second-stage mixed-task training omits in-batch negatives to avoid class false negatives. Public embedding-training data does not mean the pretrained LLM was trained from scratch with that data; snapshot leaderboard results are not current rankings.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction