Global PIQA - Parallel (Log-Likelihood Accuracy)
General QA · 2025-10-28
Global PIQA's parallel split presents aligned physical-commonsense questions with four answer options. This definition compares answer-option log likelihoods without requiring the model to generate a choice token.
Top models (higher is better)
| Model | Score |
|---|---|
| Llama 3.1 70B | 30.7 |
| Gemma 2 27B | 30.7 |
| Qwen2.5 72B | 29.8 |
| Gemma 3 12B PT | 29.3 |
| Qwen2.5 32B | 27.7 |
| Gemma 2 9B | 27.3 |
| Mixtral 8x7B | 25.2 |
| Llama 3.1 8B | 25.2 |
| Falcon 40B | 24.5 |
| Mistral 7B v0.1 Base | 23.5 |
| Falcon 7B | 21.8 |
| Gemma 3 270M | 21.1 |