Response correctness and completeness evaluation on Math
87F1 ScoreHALT
Evaluation Results
| Method | Links | |
|---|---|---|
| HALTLLM=Llama-3-70B, n_desired=n_capable2025.06 | 87 | |
| HALTLLM=Gemma-2-9B, n_desired=n_capable2025.06 | 79 | |
| UnchangedLLM=Llama-3-70B, n_desired=n_capable2025.06 | 75 | |
| RandomTrimLLM=Llama-3-70B, n_desired=n_capable2025.06 | 75 | |
| HALTLLM=Mistral-7B, n_desired=n_capable2025.06 | 69 | |
| UnchangedLLM=Llama-3-70B, n_desired=n_all2025.06 | 68 | |
| HALTLLM=Llama-3-70B, n_desired=n_all2025.06 | 68 | |
| Abstain (IDK)LLM=Llama-3-70B, n_desired=n_capable2025.06 | 68 | |
| HALTLLM=Llama-3-8B, n_desired=n_capable2025.06 | 66 | |
| RandomTrimLLM=Llama-3-70B, n_desired=n_all2025.06 | 64 | |
| UnchangedLLM=Gemma-2-9B, n_desired=n_capable2025.06 | 64 | |
| RandomTrimLLM=Gemma-2-9B, n_desired=n_capable2025.06 | 59 | |
| Abstain (IDK)LLM=Llama-3-70B, n_desired=n_all2025.06 | 53 | |
| UnchangedLLM=Mistral-7B, n_desired=n_capable2025.06 | 50 | |
| UnchangedLLM=Llama-3-8B, n_desired=n_capable2025.06 | 49 | |
| RandomTrimLLM=Mistral-7B, n_desired=n_capable2025.06 | 43 | |
| Abstain (IDK)LLM=Mistral-7B, n_desired=n_capable2025.06 | 43 | |
| UnchangedLLM=Gemma-2-9B, n_desired=n_all2025.06 | 42 | |
| HALTLLM=Gemma-2-9B, n_desired=n_all2025.06 | 41 | |
| Abstain (IDK)LLM=Llama-3-8B, n_desired=n_capable2025.06 | 38 | |
| Abstain (IDK)LLM=Gemma-2-9B, n_desired=n_capable2025.06 | 38 | |
| HALTLLM=Llama-3-8B, n_desired=n_all2025.06 | 35 | |
| UnchangedLLM=Mistral-7B, n_desired=n_all2025.06 | 31 | |
| UnchangedLLM=Llama-3-8B, n_desired=n_all2025.06 | 30 | |
| HALTLLM=Mistral-7B, n_desired=n_all2025.06 | 28 | |
| RandomTrimLLM=Gemma-2-9B, n_desired=n_all2025.06 | 28 | |
| RandomTrimLLM=Llama-3-8B, n_desired=n_capable2025.06 | 26 | |
| Abstain (IDK)LLM=Gemma-2-9B, n_desired=n_all2025.06 | 13 | |
| Abstain (IDK)LLM=Llama-3-8B, n_desired=n_all2025.06 | 12 | |
| RandomTrimLLM=Mistral-7B, n_desired=n_all2025.06 | 12 | |
| Abstain (IDK)LLM=Mistral-7B, n_desired=n_all2025.06 | 9 | |
| RandomTrimLLM=Llama-3-8B, n_desired=n_all2025.06 | 8 |