zephyr-7b-dpo-full-prometheus_consistent-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: 0.5201
• Rewards/chosen: -0.8222
• Rewards/rejected: -1.7957
• Rewards/accuracies: 0.7759
• Rewards/margins: 0.9734
• Logps/rejected: -398.6412
• Logps/chosen: -357.8258
• Logits/rejected: 1.5724
• Logits/chosen: -0.0130

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
0.5995	0.2286	100	0.5960	-0.3095	-0.7496	0.7155	0.4400	-294.0323	-306.5585	-2.3479	-2.4655
0.5431	0.4571	200	0.5326	-0.6273	-1.5233	0.7629	0.8960	-371.4070	-338.3311	-0.0959	-1.0624
0.5138	0.6857	300	0.5242	-0.5729	-1.4989	0.7629	0.9260	-368.9617	-332.8911	0.6740	-0.7677
0.5493	0.9143	400	0.5201	-0.8222	-1.7957	0.7759	0.9734	-398.6412	-357.8258	1.5724	-0.0130

Framework versions

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