results

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: 1.0888
• Rewards/chosen: -0.0906
• Rewards/rejected: -0.1196
• Rewards/accuracies: 0.8000
• Rewards/margins: 0.0290
• Logps/rejected: -1.1960
• Logps/chosen: -0.9064
• Logits/rejected: -1.3083
• Logits/chosen: -1.0410
• Nll Loss: 1.0376
• Log Odds Ratio: -0.5127
• Log Odds Chosen: 0.4766

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
1.0958	0.2020	25	1.4462	-0.1303	-0.1689	0.8000	0.0386	-1.6887	-1.3025	-1.1831	-0.9153	1.3976	-0.4854	0.5211
1.2563	0.4040	50	1.2116	-0.1057	-0.1387	0.8000	0.0330	-1.3872	-1.0575	-1.2714	-1.0083	1.1616	-0.5001	0.4913
1.3121	0.6061	75	1.1251	-0.0952	-0.1249	0.9000	0.0297	-1.2491	-0.9524	-1.3022	-1.0390	1.0740	-0.5109	0.4726
1.3689	0.8081	100	1.0888	-0.0906	-0.1196	0.8000	0.0290	-1.1960	-0.9064	-1.3083	-1.0410	1.0376	-0.5127	0.4766

Framework versions

• PEFT 0.10.1.dev0
• Transformers 4.41.0.dev0
• Pytorch 2.2.2+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1