Falcon-7B-Instruct-ORPO

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

• Loss: 1.5155
• Rewards/chosen: -0.1444
• Rewards/rejected: -0.1539
• Rewards/accuracies: 0.5090
• Rewards/margins: 0.0095
• Logps/rejected: -1.5389
• Logps/chosen: -1.4440
• Logits/rejected: -14.5432
• Logits/chosen: -14.4665
• Sft Loss: 1.4440
• Odds Ratio Loss: 0.7143

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
1.6309	0.8891	500	1.5816	-0.1510	-0.1599	0.4940	0.0089	-1.5988	-1.5096	-14.4968	-14.4213	1.5096	0.7192
1.5401	1.7782	1000	1.5269	-0.1455	-0.1549	0.5020	0.0094	-1.5492	-1.4555	-14.5486	-14.4721	1.4555	0.7147
1.4914	2.6673	1500	1.5155	-0.1444	-0.1539	0.5090	0.0095	-1.5389	-1.4440	-14.5432	-14.4665	1.4440	0.7143

Framework versions

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