zephyr-7b-dpo-full

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.4983
• Rewards/chosen: -2.4880
• Rewards/rejected: -3.6063
• Rewards/accuracies: 0.7695
• Rewards/margins: 1.1182
• Logps/rejected: -623.3074
• Logps/chosen: -511.4043
• Logits/rejected: 0.0233
• Logits/chosen: -0.4369

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
0.5875	0.21	100	0.5814	-0.6485	-1.1103	0.6953	0.4618	-373.7126	-327.4548	-1.8929	-1.8392
0.5306	0.42	200	0.5258	-1.1476	-1.9595	0.7578	0.8118	-458.6297	-377.3649	-0.1647	-0.4835
0.5097	0.63	300	0.5079	-2.3601	-3.3817	0.7656	1.0216	-600.8517	-498.6086	-0.0574	-0.4658
0.4906	0.84	400	0.5000	-2.3681	-3.4811	0.7695	1.1129	-610.7911	-499.4172	-0.0390	-0.5081

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.2
• Datasets 2.14.6
• Tokenizers 0.15.2