AWhat changed
Combined loss-balancing changes, multi-token prediction, FP8 training and pipeline communication overlap in a large mixture-of-experts model.
Scientific scope
Combined loss-balancing changes, multi-token prediction, FP8 training and pipeline communication overlap in a large mixture-of-experts model. MLA and DeepSeekMoE are inherited from earlier work, not new V3 inventions. Reported training GPU hours are not the entire research project's cost. Pinned revision appeared in 2025; original release was 2024.
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 and DeepSeekMoE are inherited from earlier work, not new V3 inventions. Reported training GPU hours are not the entire research project's cost. Pinned revision appeared in 2025; original release was 2024.
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 Technical Report
- DeepSeek-AI: DeepSeek-V3 efficient mixture-of-experts training system — official research record
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