AWhat changed
Added a low-cost lightning indexer with top-k latent KV selection and staged dense-to-sparse continued training, alongside scalable reasoning/tool-use training and synthetic interactive tasks.
Scientific scope
Added a low-cost lightning indexer with top-k latent KV selection and staged dense-to-sparse continued training, alongside scalable reasoning/tool-use training and synthetic interactive tasks. MLA/MQA and prior RL are inherited. Sparse main attention does not eliminate the indexer's pairwise scoring cost or guarantee equivalence to dense attention. September experimental and December full releases share DSA; Speciale benchmark claims are not official competition participation.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. MLA/MQA and prior RL are inherited. Sparse main attention does not eliminate the indexer's pairwise scoring cost or guarantee equivalence to dense attention. September experimental and December full releases share DSA; Speciale benchmark claims are not official competition participation.
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.
- DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models
- DeepSeek-AI: lightning-indexed sparse attention and agentic training — author/institution implementation
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