Gemma-7B-It-ORPO

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

• Loss: 1.3471
• Rewards/chosen: -0.1281
• Rewards/rejected: -0.1500
• Rewards/accuracies: 0.5610
• Rewards/margins: 0.0219
• Logps/rejected: -1.5004
• Logps/chosen: -1.2814
• Logits/rejected: 254.6614
• Logits/chosen: 254.4679
• Sft Loss: 1.2814
• Odds Ratio Loss: 0.6571

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
1.5041	0.8891	500	1.4185	-0.1352	-0.1564	0.5530	0.0212	-1.5644	-1.3522	250.7549	250.6463	1.3522	0.6626
1.428	1.7782	1000	1.3595	-0.1294	-0.1509	0.5600	0.0215	-1.5091	-1.2937	254.1350	253.9581	1.2937	0.6586
1.3302	2.6673	1500	1.3471	-0.1281	-0.1500	0.5610	0.0219	-1.5004	-1.2814	254.6614	254.4679	1.2814	0.6571

Framework versions

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