List your company

General Language Understanding on ARC-C, CSQA, HellaSwag, LAMBADA, MMLU, OpenBookQA, PIQA, Winogrande (test)

59.5ARC-C AccuracyOriginal
21.43631.31841.251.082Jun 18, 2024

Evaluation results

MethodLinks
OriginalBackbone=Phi-3.5-Mini-Instruct2024.0659.575.358.865.168.737.68074.665
OPT-OUTBackbone=Phi-3.5-Mini-Instruct2024.0658.972.957.658.26836.980.474.163.4
NPOBackbone=Phi-3.5-Mini-Instruct, Retention Training=true2024.0658.172.857.657.767.937.18074.463.2
DPOBackbone=Phi-3.5-Mini-Instruct, Retention Training=true2024.0657.972.857.655.66837.380.77563.1
IDKBackbone=Phi-3.5-Mini-Instruct, Retention Training=true2024.0657.174.357.55567.737.780.27663.2
GABackbone=Phi-3.5-Mini-Instruct, Retention Training=true2024.0656.969.456.756.867.235.879.973.162
IDKBackbone=Phi-3.5-Mini-Instruct, Collapsed=true2024.0653.770.954.847.865.736.679.476.760.7
GuardrailBackbone=Phi-3.5-Mini-Instruct2024.0652.272.959.264.167.738.378.174.563.4
OriginalBackbone=Llama-3.1-8B-Instruct2024.0651.877.159.273.268.133.880.274.164.7
NPOBackbone=Llama-3.1-8B-Instruct, Retention Training=true2024.0650.173.558.771.762.533.379.77362.8
GuardrailBackbone=Llama-3.1-8B-Instruct2024.0649.875.558.972.367.233.280.174.263.9
OPT-OUTBackbone=Llama-3.1-8B-Instruct2024.0649.875.35971.863.234.279.773.163.3
IDKBackbone=Llama-3.1-8B-Instruct, Retention Training=true2024.0649.473.858.769.763.23479.873.462.8
DPOBackbone=Llama-3.1-8B-Instruct, Retention Training=true2024.0649.268.858.768.657.933.679.872.761.1
GABackbone=Llama-3.1-8B-Instruct, Retention Training=true2024.0647.47157.771.160.732.979.272.261.5
DPOBackbone=Phi-3.5-Mini-Instruct, Collapsed=true2024.0638.936.156.754.35436.463.972.751.6
IDKBackbone=Llama-3.1-8B-Instruct, Collapsed=true2024.0638.467.653.34861.83077.171.356
NPOBackbone=Phi-3.5-Mini-Instruct, Collapsed=true2024.0628.619.656.85726.635.859.169.444.1
NPOBackbone=Llama-3.1-8B-Instruct, Collapsed=true2024.0624.72337.620.836.330.660.167.237.5
GABackbone=Phi-3.5-Mini-Instruct2024.062419.852.14723.632.653.86639.8
GABackbone=Llama-3.1-8B-Instruct, Collapsed=true2024.0623.621.932.211.633.92858.360.933.8
DPOBackbone=Llama-3.1-8B-Instruct, Collapsed=true2024.0622.930.552.836.437.928.358.469.542.1