BWhat changed
Adapted bidirectional ALiBi and staged embedding training for longer documents in v2; v3 used an XLM-RoBERTa backbone with rotary positions, task-selected LoRA adapters and dimension-flexible representation training.
Scientific scope
Adapted bidirectional ALiBi and staged embedding training for longer documents in v2; v3 used an XLM-RoBERTa backbone with rotary positions, task-selected LoRA adapters and dimension-flexible representation training. Version-specific architectures and author sets retained; v3 uses RoPE rather than v2's ALiBi. LoRA, rotary positions and Matryoshka learning are inherited tools. Shorter embedding dimensions trade quality for storage, despite the abstract's broad no-compromise wording. Longer accepted inputs are not certified lossless semantic compression.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Version-specific architectures and author sets retained; v3 uses RoPE rather than v2's ALiBi. LoRA, rotary positions and Matryoshka learning are inherited tools. Shorter embedding dimensions trade quality for storage, despite the abstract's broad no-compromise wording. Longer accepted inputs are not certified lossless semantic compression.
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.
- Jina Embeddings 2: 8192-Token General-Purpose Text Embeddings for Long Documents
- jina-embeddings-v3: Multilingual Embeddings With Task LoRA
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