phi-2-gpo-renew2-b0.001-v4-i1

This model is a fine-tuned version of DUAL-GPO/phi-2-gpo-renew2-b0.001-i0 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.0536
• Rewards/chosen: -0.0036
• Rewards/rejected: -0.0039
• Rewards/accuracies: 0.4695
• Rewards/margins: 0.0002
• Logps/rejected: -371.0876
• Logps/chosen: -399.9150
• Logits/rejected: -0.7623
• Logits/chosen: -0.8574

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
• gradient_accumulation_steps: 4
• total_train_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
0.1203	0.32	100	0.0537	-0.0024	-0.0024	0.4555	0.0001	-369.6694	-398.6797	-0.7167	-0.8167
0.1671	0.64	200	0.0537	-0.0036	-0.0037	0.4670	0.0001	-370.9240	-399.8586	-0.7745	-0.8674
0.1393	0.96	300	0.0536	-0.0038	-0.0040	0.4625	0.0003	-371.2791	-400.0731	-0.7820	-0.8772

Framework versions

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