zephyr-7b-dpo-full_lr5e-6

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

• Loss: 0.4945
• Rewards/chosen: -2.8495
• Rewards/rejected: -3.9547
• Rewards/accuracies: 0.7695
• Rewards/margins: 1.1052
• Logps/rejected: -657.4872
• Logps/chosen: -547.3602
• Logits/rejected: 4.2623
• Logits/chosen: 3.2552

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-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• total_eval_batch_size: 64
• 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	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.605	0.21	100	0.5799	-1.9111	-2.5875	0.7461	0.6764	-520.7654	-453.5230	-2.4063	-2.5964
0.5303	0.42	200	0.5347	-2.5277	-3.3091	0.7656	0.7814	-592.9240	-515.1846	2.6876	1.5286
0.5147	0.63	300	0.5106	-2.4842	-3.4691	0.75	0.9850	-608.9282	-510.8301	2.5319	1.2911
0.498	0.84	400	0.4958	-2.9252	-4.0433	0.7734	1.1181	-666.3457	-554.9301	4.6654	3.6821

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.1