zephyr-7b-dpo-full_lr1e-7

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

• Loss: 0.5999
• Rewards/chosen: -0.1059
• Rewards/rejected: -0.3888
• Rewards/accuracies: 0.7266
• Rewards/margins: 0.2829
• Logps/rejected: -300.8914
• Logps/chosen: -273.0010
• Logits/rejected: -2.2473
• Logits/chosen: -2.2980

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: 1e-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
0.6773	0.21	100	0.6767	0.0281	-0.0077	0.6914	0.0358	-262.7869	-259.5997	-2.3569	-2.4039
0.6286	0.42	200	0.6292	-0.0536	-0.2320	0.7109	0.1784	-285.2149	-267.7724	-2.3535	-2.4023
0.6161	0.63	300	0.6066	-0.0848	-0.3355	0.7188	0.2507	-295.5617	-270.8907	-2.2759	-2.3264
0.5908	0.84	400	0.6002	-0.1035	-0.3846	0.7227	0.2811	-300.4714	-272.7594	-2.2519	-2.3026

Framework versions

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