| question-answering-on-triviaqa | DeepSeek-V2 Chat (RL) (5-shot) | #1 | EM: 86.7 |
| instruction-following-on-mtbench | DeepSeek-V2 Chat (RL) | #4 | Score (1-10): 8.97 |
| instruction-following-on-alignbench | DeepSeek-V2 Chat (RL) | #5 | Score (0-10): 7.91 |
| question-answering-on-natural-questions | DeepSeek-V2 Chat (RL) (5-shot) | #8 | EM: 53.4 |
| multi-task-language-understanding-on-mmlu | DeepSeek-V2 Chat (RL) (5-shot) | #13 | Average (%): 77.8 |
| instruction-following-on-ifeval | DeepSeek-V2 Chat (RL) (0-shot) | #18 | Prompt-level loose-accuracy: 63.8 |
| arithmetic-reasoning-on-gsm8k | DeepSeek-V2 Chat (RL) (8-shot) | #51 | Accuracy: 92.2 |
| code-generation-on-mbpp | DeepSeek-V2 Chat (RL) (3-shot) | #51 | Accuracy: 72 |
| code-generation-on-humaneval | DeepSeek-V2 Chat (RL) (0-shot) | #54 | Pass@1: 81.1 |
| math-word-problem-solving-on-math | DeepSeek-V2 Chat (RL) (4-shot) | #72 | Accuracy: 53.9 |