zephyr-7b-uf-dpo-2e

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: 0.5017
• Rewards/chosen: -1.6430
• Rewards/rejected: -2.9109
• Rewards/accuracies: 0.7891
• Rewards/margins: 1.2679
• Logps/rejected: -553.7540
• Logps/chosen: -426.9287
• Logits/rejected: 2.3524
• Logits/chosen: 1.4100

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: 4
• total_train_batch_size: 256
• 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: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5424	0.5021	120	0.5378	-0.5794	-1.3437	0.7578	0.7642	-397.0288	-320.5744	-0.5261	-0.7461
0.4857	1.0042	240	0.5071	-0.9384	-1.8536	0.7695	0.9153	-448.0264	-356.4662	0.8934	0.1618
0.3605	1.5063	360	0.4996	-1.5624	-2.7607	0.7734	1.1983	-538.7272	-418.8674	2.1269	1.2559

Framework versions

• Transformers 4.44.1
• Pytorch 2.1.2+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1