Llama-2-7b-hf-eval_threapist-ORPO-filtered-0.2-version-1

This model is a fine-tuned version of meta-llama/Llama-2-7b-hf on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5471
• Rewards/chosen: -0.1206
• Rewards/rejected: -0.1280
• Rewards/accuracies: 0.6500
• Rewards/margins: 0.0074
• Logps/rejected: -1.2803
• Logps/chosen: -1.2064
• Logits/rejected: -1.4351
• Logits/chosen: -1.4353
• Nll Loss: 0.4721
• Log Odds Ratio: -0.7502
• Log Odds Chosen: 0.0744

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: 2
• total_train_batch_size: 4
• 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
0.8049	0.6	141	0.8303	-0.2121	-0.2243	0.6500	0.0122	-2.2432	-2.1211	-0.7703	-0.7698	0.7515	-0.7874	0.1083
0.5299	1.2	282	0.5796	-0.1515	-0.1597	0.6500	0.0081	-1.5968	-1.5153	-1.2952	-1.2957	0.5038	-0.7583	0.0737
0.595	1.8	423	0.5535	-0.1227	-0.1296	0.6500	0.0070	-1.2962	-1.2265	-1.4015	-1.4015	0.4782	-0.7532	0.0683
0.6024	2.4	564	0.5476	-0.1216	-0.1293	0.6500	0.0077	-1.2929	-1.2164	-1.4239	-1.4238	0.4727	-0.7491	0.0768
0.59	3.0	705	0.5471	-0.1206	-0.1280	0.6500	0.0074	-1.2803	-1.2064	-1.4351	-1.4353	0.4721	-0.7502	0.0744

Framework versions

• PEFT 0.10.0
• Transformers 4.40.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1