zephyr-7b-dpo-qlora

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

• Loss: 0.6813
• Rewards/chosen: -0.0009
• Rewards/rejected: -0.0252
• Rewards/accuracies: 0.2920
• Rewards/margins: 0.0243
• Logps/rejected: -71.3009
• Logps/chosen: -65.4449
• Logits/rejected: -2.4428
• Logits/chosen: -2.4444

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
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• 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
0.69	0.26	100	0.6897	0.0232	0.0168	0.2680	0.0064	-67.1001	-63.0342	-2.4904	-2.4911
0.6869	0.52	200	0.6849	0.0066	-0.0092	0.3060	0.0159	-69.7060	-64.6950	-2.4556	-2.4573
0.681	0.78	300	0.6815	-0.0026	-0.0264	0.2880	0.0238	-71.4280	-65.6224	-2.4430	-2.4446

Framework versions

• PEFT 0.7.1
• Transformers 4.36.2
• Pytorch 2.1.2+cu118
• Datasets 2.14.6
• Tokenizers 0.15.2