Correctness Prediction on PIQA
80.69AccuracyGRIFFIN
Evaluation results
| Method | Links | |
|---|---|---|
| GRIFFINBackbone=Mistral 7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 80.69 | |
| I-GLASSBackbone=Mistral 7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 80.47 | |
| I-GLASSBackbone=Gemma 7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 80.25 | |
| GRIFFINBackbone=Gemma 7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 80.03 | |
| KMEANSTrain samples=10242026.01 | 79.64 | |
| KMEANSTrain samples=5122026.01 | 79.32 | |
| kNNTrain samples=5122026.01 | 78.38 | |
| I-GLASSBackbone=Llama2 7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 78.07 | |
| GRIFFINBackbone=Llama2 7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 78.07 | |
| IRT-NETTrain samples=10242026.01 | 77.43 | |
| I-GLASSBackbone=ReLU-Llama2 7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 77.42 | |
| GRIFFINBackbone=ReLU-Llama2 7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 77.42 | |
| EMBEDLLMTrain samples=5122026.01 | 76.86 | |
| EMBEDLLMTrain samples=10242026.01 | 76.61 | |
| kNNTrain samples=10242026.01 | 76.32 | |
| I-GLASSBackbone=OPT 6.7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 75.9 | |
| GRIFFINBackbone=OPT 6.7B, Sparsity=50% FF, Evaluation Protocol=0-shot unnormalized2025.08 | 75.73 | |
| IRT-NETTrain samples=5122026.01 | 75.69 | |
| EMBEDLLM+Train samples=10242026.01 | 74.41 | |
| LOCUSTrain samples=10242026.01 | 73.03 | |
| LOCUSTrain samples=2562026.01 | 72.21 | |
| EMBEDLLM+Train samples=5122026.01 | 68.97 | |
| LOCUSTrain samples=5122026.01 | 65.78 | |
| KMEANSTrain samples=2562026.01 | 65.51 | |
| IRT-NETTrain samples=2562026.01 | 57.39 | |
| EMBEDLLMTrain samples=2562026.01 | 57.24 | |
| EMBEDLLM+Train samples=2562026.01 | 47.88 | |
| kNNTrain samples=2562026.01 | 26.33 |