OrpoLlama-3-8B-Instruct

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.0517
• Rewards/chosen: -0.0621
• Rewards/rejected: -0.0634
• Rewards/accuracies: 0.6000
• Rewards/margins: 0.0013
• Logps/rejected: -0.6340
• Logps/chosen: -0.6207
• Logits/rejected: -0.2825
• Logits/chosen: -0.2736
• Nll Loss: 0.9848
• Log Odds Ratio: -0.6691
• Log Odds Chosen: 0.1127

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: 3

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
2.2564	0.5980	74	1.8521	-0.1306	-0.1270	0.6000	-0.0036	-1.2696	-1.3055	-0.4912	-0.5850	1.7776	-0.7454	-0.0391
1.8749	1.1960	148	1.3145	-0.0855	-0.0879	0.6000	0.0024	-0.8795	-0.8553	-0.2945	-0.2832	1.2482	-0.6628	0.1091
1.1933	1.7939	222	1.1033	-0.0662	-0.0667	0.6000	0.0005	-0.6671	-0.6624	-0.2268	-0.2101	1.0354	-0.6787	0.0828
0.8761	2.3919	296	1.0517	-0.0621	-0.0634	0.6000	0.0013	-0.6340	-0.6207	-0.2825	-0.2736	0.9848	-0.6691	0.1127

Framework versions

• PEFT 0.11.1
• Transformers 4.41.2
• Pytorch 2.3.1+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1