llama3-1-ox-llms-8b-dpo-full

This model is a fine-tuned version of oxford-llms/llama3-1-ox-llms-8b-sft-full on the argilla/dpo-mix-7k dataset. It achieves the following results on the evaluation set:

• Loss: 0.4831
• Rewards/chosen: -0.2543
• Rewards/rejected: -1.0940
• Rewards/accuracies: 0.7708
• Rewards/margins: 0.8397
• Logps/rejected: -340.3136
• Logps/chosen: -325.1967
• Logits/rejected: -1.3101
• Logits/chosen: -1.3085

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• 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
0.528	0.9479	100	0.5267	-0.0255	-0.6055	0.7604	0.5800	-330.5430	-320.6201	-1.3169	-1.3159
0.3731	1.8957	200	0.4821	-0.2481	-1.0900	0.7604	0.8419	-340.2323	-325.0733	-1.3099	-1.3082

Framework versions

• Transformers 4.45.1
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3