orpo-model-output

This model is a fine-tuned version of openai-community/gpt2 on the piqa dataset. It achieves the following results on the evaluation set:

• Loss: 3.2999
• Rewards/chosen: -0.3182
• Rewards/rejected: -0.3369
• Rewards/accuracies: 0.6693
• Rewards/margins: 0.0187
• Logps/rejected: -3.3688
• Logps/chosen: -3.1821
• Logits/rejected: -24.5739
• Logits/chosen: -24.6603
• Nll Loss: 3.2330
• Log Odds Ratio: -0.6695
• Log Odds Chosen: 0.2013

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Nll Loss	Log Odds Ratio	Log Odds Chosen
3.1799	1.0	3223	3.1575	-0.2992	-0.3103	0.6472	0.0111	-3.1030	-2.9917	-16.2781	-16.3180	3.0900	-0.6758	0.1201
2.7077	2.0	6446	3.1544	-0.3005	-0.3160	0.6652	0.0154	-3.1595	-3.0051	-21.7517	-21.8387	3.0878	-0.6671	0.1676
2.2691	3.0	9669	3.2999	-0.3182	-0.3369	0.6693	0.0187	-3.3688	-3.1821	-24.5739	-24.6603	3.2330	-0.6695	0.2013

Framework versions

• Transformers 4.44.2
• Pytorch 2.0.0
• Datasets 2.16.1
• Tokenizers 0.19.1