dpo-selective-buffer-spo-shift

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

• Loss: 2897.3179
• Rewards/chosen: -0.6441
• Rewards/rejected: -0.5754
• Rewards/accuracies: 0.4239
• Rewards/margins: -0.0687
• Rewards/safe Rewards: -0.6417
• Rewards/unsafe Rewards: -0.6386
• Logps/rejected: -150.0089
• Logps/chosen: -194.8470
• Logits/rejected: -1.5786
• Logits/chosen: -1.8526

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: 2
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• 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
4952.3219	0.27	500	3031.1638	-0.6585	-0.5935	0.4215	-0.0650	-0.6552	-0.6514	-151.8219	-196.2862	-1.6812	-1.9554
4670.3188	0.54	1000	2931.8032	-0.6547	-0.5940	0.4353	-0.0607	-0.6514	-0.6477	-151.8720	-195.9055	-1.5084	-1.7952
4368.5492	0.81	1500	2899.6733	-0.6504	-0.5839	0.4268	-0.0665	-0.6481	-0.6452	-150.8582	-195.4787	-1.5552	-1.8340

Framework versions

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