merged_model_dpo

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0000
• Rewards/chosen: 0.2797
• Rewards/rejected: -17.5881
• Rewards/accuracies: 1.0
• Rewards/margins: 17.8678
• Logps/rejected: -299.3185
• Logps/chosen: -28.7786
• Logits/rejected: -3.7935
• Logits/chosen: -4.0383

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: 0.0002
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2
• training_steps: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6218	0.21	10	0.2547	0.5225	-0.8859	1.0	1.4085	-132.2969	-26.3501	-3.7422	-3.8380
0.1216	0.43	20	0.0055	0.6847	-5.7740	1.0	6.4587	-181.1776	-24.7284	-3.7620	-3.9325
0.0074	0.64	30	0.0000	0.4694	-13.1598	1.0	13.6292	-255.0354	-26.8815	-3.7881	-4.0116
0.0001	0.85	40	0.0000	0.3177	-16.7606	1.0	17.0783	-291.0435	-28.3980	-3.7933	-4.0344
0.0001	1.06	50	0.0000	0.2797	-17.5881	1.0	17.8678	-299.3185	-28.7786	-3.7935	-4.0383

Framework versions

• PEFT 0.7.2.dev0
• Transformers 4.37.0.dev0
• Pytorch 2.1.0+cu118
• Datasets 2.16.0
• Tokenizers 0.15.0