Falcon-7B-Instruct-ORPO-SALT-HALF

This model is a fine-tuned version of tiiuae/falcon-7b-instruct on the dpo_mix_en and the bct_non_cot_dpo_500 datasets. It achieves the following results on the evaluation set:

• Loss: 1.4967
• Rewards/chosen: -0.1424
• Rewards/rejected: -0.1512
• Rewards/accuracies: 0.5038
• Rewards/margins: 0.0088
• Logps/rejected: -1.5123
• Logps/chosen: -1.4238
• Logits/rejected: -14.4005
• Logits/chosen: -14.4333
• Sft Loss: 1.4238
• Odds Ratio Loss: 0.7294

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: 5e-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Sft Loss	Odds Ratio Loss
1.5988	0.8467	500	1.5655	-0.1492	-0.1572	0.4962	0.0080	-1.5724	-1.4919	-14.3530	-14.3859	1.4919	0.7361
1.4213	1.6935	1000	1.5097	-0.1437	-0.1524	0.5038	0.0087	-1.5240	-1.4366	-14.3997	-14.4324	1.4366	0.7303
1.4234	2.5402	1500	1.4967	-0.1424	-0.1512	0.5038	0.0088	-1.5123	-1.4238	-14.4005	-14.4333	1.4238	0.7294

Framework versions

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