Mistral-7B-Instruct-v0.2-ORPO-SALT-HALF

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on the dpo_mix_en and the bct_non_cot_dpo_500 datasets. It achieves the following results on the evaluation set:

• Loss: 0.8831
• Rewards/chosen: -0.0820
• Rewards/rejected: -0.1062
• Rewards/accuracies: 0.5810
• Rewards/margins: 0.0242
• Logps/rejected: -1.0619
• Logps/chosen: -0.8200
• Logits/rejected: -2.8746
• Logits/chosen: -2.8587
• Sft Loss: 0.8200
• Odds Ratio Loss: 0.6316

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
0.9115	0.8467	500	0.9161	-0.0853	-0.1100	0.5733	0.0247	-1.0998	-0.8530	-2.8628	-2.8480	0.8530	0.6317
0.8308	1.6935	1000	0.8892	-0.0826	-0.1068	0.5724	0.0242	-1.0678	-0.8261	-2.8569	-2.8411	0.8261	0.6311
0.8476	2.5402	1500	0.8831	-0.0820	-0.1062	0.5810	0.0242	-1.0619	-0.8200	-2.8746	-2.8587	0.8200	0.6316

Framework versions

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