zephyr-7b-dpo-full-prometheus-3

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4961
• Rewards/chosen: -1.0197
• Rewards/rejected: -1.9078
• Rewards/accuracies: 0.7241
• Rewards/margins: 0.8881
• Logps/rejected: -439.0560
• Logps/chosen: -361.9284
• Logits/rejected: 2.1814
• Logits/chosen: 1.1464

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: 55
• 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
0.5621	0.2286	100	0.5630	-0.6661	-1.2098	0.6422	0.5437	-369.2566	-326.5674	-0.4908	-0.7741
0.5053	0.4571	200	0.5191	-0.9978	-1.7738	0.7241	0.7760	-425.6610	-359.7413	1.7795	1.0550
0.4913	0.6857	300	0.5001	-1.0485	-1.9070	0.7328	0.8584	-438.9752	-364.8153	2.0188	0.9765
0.5017	0.9143	400	0.4961	-1.0197	-1.9078	0.7241	0.8881	-439.0560	-361.9284	2.1814	1.1464

Framework versions

• Transformers 4.44.0.dev0
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1