Reference library
DeepSeek-V3 Technical Report
Paper summary
DeepSeek-V3 develops a large mixture-of-experts model with latent attention, load-balancing changes, and multi-token prediction. The report describes pretraining and post-training and presents task benchmarks alongside training-resource accounting.
Relevance to the project
Provides examples of sparse execution, load balancing, and explicit compute records. These may inform future GPU benchmarking without selecting an MoE simulation architecture.
Reported total and active parameters are different quantities.
Source record
- Reference ID
- deepseek-v3
- Archived version
- 2412.19437v2
- Summary provenance
- reference-library/papers/deepseek-v3/metadata.json · summary
- Review status
- initial orientation summary; not a critical full-paper review
- Reading scope
- Selected abstract/opening and available closing discussion/conclusion passages in the preserved text extract; broader project notes provide context. No reproduction of experiments, exhaustive methods review, or equation/figure audit.