phi3-offline-dpo-lora-noise-0.0-5e-6-42

This model is a fine-tuned version of microsoft/Phi-3-mini-4k-instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6639
• Rewards/chosen: -0.1248
• Rewards/rejected: -0.1933
• Rewards/accuracies: 0.7421
• Rewards/margins: 0.0685
• Logps/rejected: -403.0595
• Logps/chosen: -421.0499
• Logits/rejected: 12.1072
• Logits/chosen: 13.9087

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6931	0.1778	100	0.6835	-0.0511	-0.0728	0.6905	0.0218	-391.0186	-413.6780	12.3764	14.1803
0.689	0.3556	200	0.6682	-0.1441	-0.2014	0.7460	0.0573	-403.8743	-422.9761	12.1803	13.9841
0.6923	0.5333	300	0.6673	-0.1140	-0.1749	0.7897	0.0609	-401.2295	-419.9747	12.1769	13.9748
0.6914	0.7111	400	0.6655	-0.1195	-0.1839	0.7698	0.0644	-402.1236	-420.5240	12.1267	13.9317
0.696	0.8889	500	0.6633	-0.1262	-0.1959	0.7540	0.0697	-403.3280	-421.1901	12.0952	13.8997

Framework versions

• PEFT 0.7.1
• Transformers 4.42.3
• Pytorch 2.3.0+cu121
• Datasets 2.14.6
• Tokenizers 0.19.1