zephyr-7b-dpo-full

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.5020
• Rewards/chosen: -0.8985
• Rewards/rejected: -1.8744
• Rewards/accuracies: 0.7812
• Rewards/margins: 0.9759
• Logps/rejected: -450.1291
• Logps/chosen: -352.4258
• Logits/rejected: 1.7371
• Logits/chosen: 0.9003

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-07
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• total_eval_batch_size: 64
• 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
0.5709	0.2092	100	0.5765	-0.3415	-0.8321	0.7305	0.4906	-345.8958	-296.7220	-1.0885	-1.2465
0.5427	0.4184	200	0.5256	-0.7352	-1.5375	0.7695	0.8022	-416.4311	-336.0986	0.9059	0.1815
0.4892	0.6276	300	0.5082	-0.8910	-1.8210	0.7695	0.9300	-444.7822	-351.6719	1.3892	0.5828
0.5037	0.8368	400	0.5031	-0.8365	-1.7881	0.7852	0.9517	-441.4968	-346.2211	1.6106	0.7959

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0
• Datasets 3.1.0
• Tokenizers 0.19.1