Memory Analysis on the Training Course of DeepSeek Models
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916608770834432 |
|---|---|
| author | Zhang, Ping Su, Lei |
| author_facet | Zhang, Ping Su, Lei |
| contents | We present a theoretical analysis of GPU memory consumption during the training of DeepSeek models such as DeepSeek-v2 and DeepSeek-v3. Our primary objective is to clarify the device-level memory requirements associated with various distributed training configurations. Specifically, we examine critical factors influencing memory usage, including micro-batch size, activation recomputation policies, 3D parallelism, and ZeRO optimizations. It is important to emphasize that the training policies discussed in this report are not representative of DeepSeek's official configurations. Instead, they are explored to provide a deeper understanding of memory dynamics in training of large-scale mixture-of-experts model. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_07846 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Memory Analysis on the Training Course of DeepSeek Models Zhang, Ping Su, Lei Performance Machine Learning We present a theoretical analysis of GPU memory consumption during the training of DeepSeek models such as DeepSeek-v2 and DeepSeek-v3. Our primary objective is to clarify the device-level memory requirements associated with various distributed training configurations. Specifically, we examine critical factors influencing memory usage, including micro-batch size, activation recomputation policies, 3D parallelism, and ZeRO optimizations. It is important to emphasize that the training policies discussed in this report are not representative of DeepSeek's official configurations. Instead, they are explored to provide a deeper understanding of memory dynamics in training of large-scale mixture-of-experts model. |
| title | Memory Analysis on the Training Course of DeepSeek Models |
| topic | Performance Machine Learning |
| url | https://arxiv.org/abs/2502.07846 |