llama-3-instruct-8b-orpo-full

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

• Loss: 0.6673
• Rewards/chosen: -0.0028
• Rewards/rejected: -0.0031
• Rewards/accuracies: 0.5474
• Rewards/margins: 0.0002
• Logps/rejected: -0.3050
• Logps/chosen: -0.2820
• Logits/rejected: -1.4120
• Logits/chosen: -1.4466
• Nll Loss: 0.6694
• Log Odds Ratio: -0.6793
• Log Odds Chosen: 0.0964

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-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• 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: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Nll Loss	Log Odds Ratio	Log Odds Chosen
0.6876	0.9174	400	0.6673	-0.0028	-0.0031	0.5474	0.0002	-0.3050	-0.2820	-1.4120	-1.4466	0.6694	-0.6793	0.0964

Framework versions

• Transformers 4.44.2
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.0