reward-judge_iter-dpo-genRM_pilot-exp_iter3

This model is a fine-tuned version of RyanYr/reward-judge_iter-dpo-genRM_pilot-exp_iter2 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2888
• Rewards/chosen: 2.8310
• Rewards/rejected: 0.6403
• Rewards/accuracies: 0.8800
• Rewards/margins: 2.1907
• Logps/rejected: -222.9084
• Logps/chosen: -232.9613
• Logits/rejected: -0.8347
• Logits/chosen: -0.7593

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
0.5718	0.9370	75	0.3895	3.7455	2.3739	0.7200	1.3716	-205.5720	-223.8162	-0.7590	-0.6813
0.3829	1.8739	150	0.2888	2.8310	0.6403	0.8800	2.1907	-222.9084	-232.9613	-0.8347	-0.7593

Framework versions

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