BWhat changed
Fine-tuned one pretrained model with bidirectional contrastive representation loss and causal next-token generation loss, switching behavior by task formatting and instructions.
Scientific scope
Fine-tuned one pretrained model with bidirectional contrastive representation loss and causal next-token generation loss, switching behavior by task formatting and instructions. Pooling excludes instruction/format tokens but attention still depends on them. Additional representation batches/objectives add training compute; task/loss balancing and inherited pretraining matter. Matching separate specialists in reported tests is not a theorem of no performance trade-off for every dataset or a universal RAG speedup.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Pooling excludes instruction/format tokens but attention still depends on them. Additional representation batches/objectives add training compute; task/loss balancing and inherited pretraining matter. Matching separate specialists in reported tests is not a theorem of no performance trade-off for every dataset or a universal RAG speedup.
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