dpo_t5

This model is a fine-tuned version of zera09/long_t5_4 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1323
• Rewards/chosen: -2.0711
• Rewards/rejected: -8.9414
• Rewards/accuracies: 0.8944
• Rewards/margins: 6.8703
• Logps/rejected: -20.2375
• Logps/chosen: -30.8509
• Logits/rejected: -17.6818
• Logits/chosen: -17.7684

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: 1e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.2665	0.625	500	0.1528	-0.7661	-5.9810	0.8906	5.2149	-15.3035	-28.6760	-18.1636	-18.1581
0.1513	1.25	1000	0.1361	-1.7646	-8.3032	0.8963	6.5386	-19.1739	-30.3402	-17.7553	-17.8268
0.1242	1.875	1500	0.1325	-2.0594	-8.8889	0.8950	6.8295	-20.1501	-30.8315	-17.6642	-17.7493
0.1112	2.5	2000	0.1323	-2.0711	-8.9414	0.8944	6.8703	-20.2375	-30.8509	-17.6818	-17.7684

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.1+cu118
• Datasets 2.20.0
• Tokenizers 0.19.1