zephyr-7b-dpo-full_sft3epoch

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

• Loss: 0.6172
• Rewards/chosen: -1.5792
• Rewards/rejected: -1.8655
• Rewards/accuracies: 0.625
• Rewards/margins: 0.2863
• Logps/rejected: -1146.1112
• Logps/chosen: -1218.1312
• Logits/rejected: -3.6422
• Logits/chosen: -3.6317

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
0.6601	0.21	100	0.6572	-0.5696	-0.7082	0.6133	0.1387	-1030.3876	-1117.1681	-3.8281	-3.8175
0.6329	0.42	200	0.6378	-1.1629	-1.3547	0.6523	0.1918	-1095.0327	-1176.4983	-3.7205	-3.7123
0.6251	0.63	300	0.6219	-1.5227	-1.7758	0.6484	0.2530	-1137.1422	-1212.4856	-3.6798	-3.6707
0.6163	0.84	400	0.6192	-1.4583	-1.7357	0.6289	0.2774	-1133.1334	-1206.0380	-3.6473	-3.6358

Framework versions

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