dpo-selective-buffer-safeipo

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

• Loss: 4449.9023
• Rewards/chosen: -0.8766
• Rewards/rejected: -0.9587
• Rewards/accuracies: 0.6161
• Rewards/margins: 0.0822
• Rewards/safe Rewards: -0.8653
• Rewards/unsafe Rewards: -0.8608
• Logps/rejected: -198.0037
• Logps/chosen: -228.0047
• Logits/rejected: 1.7482
• Logits/chosen: 0.9054

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-07
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Rewards/safe Rewards	Rewards/unsafe Rewards	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
5410.1973	0.27	500	4657.3340	-0.6508	-0.7493	0.6367	0.0984	-0.6382	-0.6354	-177.0600	-205.4323	0.6948	-0.0099
5634.6316	0.53	1000	4507.8945	-0.8000	-0.8748	0.6152	0.0748	-0.7886	-0.7846	-189.6167	-220.3491	1.1542	0.4120
5749.5141	0.8	1500	4458.4429	-0.8858	-0.9723	0.6194	0.0865	-0.8741	-0.8700	-199.3641	-228.9305	1.9547	1.0718

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2
• Datasets 2.14.6
• Tokenizers 0.15.0