SFT
×Stage · DeepSeek-V4.1-Flash
Supervised fine-tuning, the first of post-training’s three stages; the report gives it no settings.
- Place
- first of three
Not disclosed Context of this stage · What is trained · Data · Epochs · Learning rate · Tokens
Architecture at this stageStated for post-training as a whole, with no stage named: the Hierarchical Sparse Indexer is switched on, the main KV cache is trained in FP4, and DSpark trains beside the backbone with its gradient cut.
All settings6
- Place
- First of three: SFT, then RL, then on-policy distillation, “without algorithmic modifications beyond well-established practices”
- DSpark
- Trained alongside the backbone through post-training, with no gradient from its objective into the backbone
- Hierarchical Sparse Indexer
- “Training-aware and introduced in post-training”: later indexers score a candidate pool an earlier one selected
- Number format
- Quantization-aware training for the main KV cache in FP4, introduced in post-training; the sliding-window KV cache stays FP8
- Replay
- The decoder’s bounded replay of the sliding window is simulated in post-training
- Long sequences
- “A substantial share of training occurring at ultra-long sequence lengths in both the pre-training and post-training stages”, of sequences up to one million tokens
Sources5
- DeepSeek_V41_Tech_Report.pdf, p. 25Place
- DeepSeek_V41_Tech_Report.pdf, p. 14DSpark, Number format
- DeepSeek_V41_Tech_Report.pdf, p. 12Hierarchical Sparse Indexer
- DeepSeek_V41_Tech_Report.pdf, p. 20Replay
- DeepSeek_V41_Tech_Report.pdf, p. 17Long sequences