zephyr-7b-dpo-full_sft2epoch

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

• Loss: 0.6211
• Rewards/chosen: -1.4941
• Rewards/rejected: -1.7515
• Rewards/accuracies: 0.625
• Rewards/margins: 0.2573
• Logps/rejected: -1191.0941
• Logps/chosen: -1275.2760
• Logits/rejected: -3.7499
• Logits/chosen: -3.7463

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: 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
0.6596	0.21	100	0.6582	-0.3306	-0.4516	0.6172	0.1209	-1061.1038	-1158.9254	-3.8934	-3.8869
0.6329	0.42	200	0.6346	-0.8973	-1.0821	0.6172	0.1848	-1124.1610	-1215.5958	-3.8301	-3.8271
0.6282	0.63	300	0.6261	-1.3758	-1.6045	0.6172	0.2287	-1176.3937	-1263.4438	-3.7836	-3.7816
0.6216	0.84	400	0.6214	-1.5153	-1.7774	0.6562	0.2621	-1193.6827	-1277.3909	-3.7427	-3.7397

Framework versions

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