deepseek-ai
DeepSeek-V4-Flash-0731
DeepSeek V4 Flash 0731 is an official text-model release with MoE, attention compression and a DSpark speculative decoding module. Its card supersedes the preview, which remains in the inventory.
Specifications
- Source-reported parameters
- Unknown
- Architectures
- Unknown
- Layers
- 43
- KV heads
- 1
- Head dimension
- 512
- config.json context
- 1048576
- Native context in model card
- Unknown
- Extended context
- Unknown
- Weight dtype
- Unknown
- File formats
- Unknown
- Declared languages
- Unknown
- Estimation method
- Not validated