Llama-2-7b-hf-eval_threapist-ORPO-filtered-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: 0.9066
• Rewards/chosen: -0.2061
• Rewards/rejected: -0.2562
• Rewards/accuracies: 0.7000
• Rewards/margins: 0.0500
• Logps/rejected: -2.5615
• Logps/chosen: -2.0611
• Logits/rejected: -1.0229
• Logits/chosen: -1.0111
• Nll Loss: 0.8524
• Log Odds Ratio: -0.5638
• Log Odds Chosen: 0.5622

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: 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	Nll Loss	Log Odds Ratio	Log Odds Chosen
0.7757	0.4037	44	1.0979	-0.2148	-0.2646	0.7000	0.0498	-2.6460	-2.1485	-0.9189	-0.9063	1.0343	-0.5790	0.5551
0.7944	0.8073	88	1.0113	-0.2113	-0.2607	0.7000	0.0494	-2.6073	-2.1130	-0.9711	-0.9585	0.9516	-0.5747	0.5532
0.6896	1.2110	132	0.9476	-0.2082	-0.2579	0.7000	0.0497	-2.5794	-2.0824	-0.9990	-0.9867	0.8912	-0.5690	0.5574
0.8565	1.6147	176	0.9066	-0.2061	-0.2562	0.7000	0.0500	-2.5615	-2.0611	-1.0229	-1.0111	0.8524	-0.5638	0.5622

Framework versions

• PEFT 0.10.0
• Transformers 4.40.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1