reward-judge_iter-dpo-genRM_pilot-exp_iter2

This model is a fine-tuned version of RyanYr/reward-judge_iter-dpo-genRM_pilot-exp_iter1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2833
• Rewards/chosen: 0.6682
• Rewards/rejected: -1.6351
• Rewards/accuracies: 0.9200
• Rewards/margins: 2.3033
• Logps/rejected: -185.4352
• Logps/chosen: -206.2261
• Logits/rejected: -0.7883
• Logits/chosen: -0.6863

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: 2e-07
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 128
• total_train_batch_size: 512
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.4758	0.9370	75	0.5220	1.9179	0.8606	0.8400	1.0573	-160.4774	-193.7286	-0.7280	-0.6243
0.2866	1.8739	150	0.2833	0.6682	-1.6351	0.9200	2.3033	-185.4352	-206.2261	-0.7883	-0.6863

Framework versions

• Transformers 4.43.4
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.19.1