ft-Llama3-8b-orpo

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B on the mlabonne/orpo-dpo-mix-40k dataset. It achieves the following results on the evaluation set:

• Loss: 0.8983
• Rewards/chosen: -0.0999
• Rewards/rejected: -0.1748
• Rewards/accuracies: 0.4000
• Rewards/margins: 0.0749
• Logps/rejected: -1.7478
• Logps/chosen: -0.9993
• Logits/rejected: -1.5466
• Logits/chosen: -1.5315
• Nll Loss: 0.8281
• Log Odds Ratio: -0.7026
• Log Odds Chosen: 0.7314

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: 8e-06
• 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: linear
• lr_scheduler_warmup_steps: 10
• 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
1.6579	0.2	25	1.2469	-0.1560	-0.2318	0.5	0.0758	-2.3180	-1.5595	-1.2300	-1.0199	1.1776	-0.6935	0.7440
1.1014	0.4	50	1.0297	-0.1262	-0.1994	0.5	0.0732	-1.9942	-1.2621	-1.4006	-1.3743	0.9587	-0.7096	0.7137
0.9391	0.61	75	0.9463	-0.1106	-0.1844	0.5	0.0738	-1.8440	-1.1062	-1.5970	-1.5504	0.8754	-0.7083	0.7185
0.676	0.81	100	0.8983	-0.0999	-0.1748	0.4000	0.0749	-1.7478	-0.9993	-1.5466	-1.5315	0.8281	-0.7026	0.7314

Framework versions

• PEFT 0.10.0
• Transformers 4.39.3
• Pytorch 2.4.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2