DeepSeek
Benchmarks
TaskDatasetSOTA resultTrendResults
DeepSeek
0NR Score
20
DeepSeek-7b five finetuned variants
3.8Average ASR
16
deepseek-7b v1 (pretrained)
100ASR (%)
13
DeepSeek-V2-Lite-Chat 15.7B
49.91Inference Time (ms)
10
DeepSeek V3.2
78.5Attack Success Rate
9
DeepSeek Paraphrased V3
0.418ROC AUC (1% FPR)
8
DeepSeek-7B
100AUC
7
DeepSeek
99.3Detection Rate
7
DeepSeek-V4-Pro generated SFT targets
0.977Entity Fidelity
6
DeepSeek-V3 primary grid (1,008 balanced runs)
3.79Quality Score
5
DeepSeek V3
4.12Compliance
5
DeepSeek-V2-Lite workload
114,600Training Throughput (tokens/s)
3
DeepSeek Lite V2
8.8Memory Usage (GB)
3
DeepSeek-V3 Weights
0Weight ΔW L2 Distance
3
DeepSeek-V2-Lite (16B) (train)
55.3Average Optimizer State Memory (MB)
2