Question Answering on Truthful QA (LIS, Accuracy)
3.1838LISDeepSeek V2 Lite
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DeepSeek V2 LiteStrategy=Raw2026.02 | 3.1838 | 0.349 | |
| DeepSeek-MoEStrategy=Raw2026.02 | 2.9672 | 0.311 | |
| DeepSeek V2 LiteStrategy=R&Q2026.02 | 2.3862 | 0.35 | |
| DeepSeek-MoEStrategy=R&Q2026.02 | 2.1749 | 0.311 | |
| Switch Transformer (16 experts)Strategy=Raw2026.02 | 2.0121 | 0.374 | |
| Switch Transformer (16 experts)Strategy=R&Q2026.02 | 1.6856 | 0.369 | |
| Switch Transformer (8 experts)Strategy=Raw2026.02 | 1.4956 | 0.369 | |
| Switch Transformer (8 experts)Strategy=R&Q2026.02 | 1.3494 | 0.364 | |
| LLaMa-MoEStrategy=Raw2026.02 | 1.2946 | 0.273 | |
| LLaMa-MoEStrategy=R&Q2026.02 | 1.2025 | 0.273 |