dpo-test

This model is a fine-tuned version of meta-llama/Llama-2-7b-chat-hf on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0060
• Rewards/chosen: -2.4578
• Rewards/rejected: -10.0480
• Rewards/accuracies: 1.0
• Rewards/margins: 7.5902
• Logps/rejected: -189.0133
• Logps/chosen: -107.4476
• Logits/rejected: -0.6723
• Logits/chosen: -0.6997

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: 0.0001
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 16
• 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: linear
• lr_scheduler_warmup_steps: 150
• 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
0.139	2.29	500	0.1416	-1.3038	-4.9329	0.9464	3.6291	-137.8629	-95.9085	-0.7743	-0.7799
0.0292	4.58	1000	0.0326	-2.0104	-7.6961	0.9974	5.6857	-165.4948	-102.9742	-0.6749	-0.6925
0.0118	6.87	1500	0.0129	-2.2528	-9.2582	1.0	7.0055	-181.1160	-105.3981	-0.6526	-0.6741
0.0056	9.17	2000	0.0060	-2.4578	-10.0480	1.0	7.5902	-189.0133	-107.4476	-0.6723	-0.6997

Framework versions

• PEFT 0.7.1
• Transformers 4.36.2
• Pytorch 2.1.2+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0