High Throughput Inference Recommendation: DeepSeek V3
High Throughput Inference → DeepSeek V3
Anwendungsfallübersicht
DeepSeek V3 by Deepseek is a recommended option for the High Throughput Inference use case.
Empfohlene Modelle
| Rang | Anbieter | Kontextfenster | Score |
|---|
| #1 | Deepseek | 64K | 37/100 |
Leistungsanalyse
Benchmark-Leistung
| Benchmark | Ergebnis |
|---|
| MMLU | 88.5 |
| HUMANEVAL | 82.6 |
| GSM8K | 89.3 |
| MATH | 61.6 |
| BBH | 84.9 |
Stärken
- MMLU score 88.5, strong knowledge reasoning.
- HumanEval 82.6, excellent code generation.
- GSM8K 89.3, robust math reasoning.
- 采用 MoE 混合专家架构。
Anforderungen
- 需要 deepseek 的 API 密钥
- 输入长度须在 64K 上下文窗口内