llama3-8b-base-dpo-120

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4893
• Rewards/chosen: -0.5883
• Rewards/rejected: -1.3409
• Rewards/accuracies: 0.6905
• Rewards/margins: 0.7525
• Logps/rejected: -293.5897
• Logps/chosen: -331.5697
• Logits/rejected: 0.4485
• Logits/chosen: 0.2338

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: 1
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 24
• gradient_accumulation_steps: 5
• total_train_batch_size: 120
• total_eval_batch_size: 48
• 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
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.467	0.4906	250	0.4987	-0.6034	-1.2547	0.7262	0.6512	-291.8655	-331.8713	0.4838	0.2542
0.4536	0.9812	500	0.4893	-0.5883	-1.3409	0.6905	0.7525	-293.5897	-331.5697	0.4485	0.2338

Framework versions

• Transformers 4.42.0
• Pytorch 2.2.2+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1