phi-2-gpo-renew2-b0.001-v2-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: 0.0539
• Rewards/chosen: 0.0016
• Rewards/rejected: 0.0019
• Rewards/accuracies: 0.4295
• Rewards/margins: -0.0003
• Logps/rejected: -365.2908
• Logps/chosen: -394.6620
• Logits/rejected: -0.7531
• Logits/chosen: -0.8487

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
0.1204	0.32	100	0.0538	0.0003	0.0005	0.4400	-0.0002	-366.7078	-395.9741	-0.7404	-0.8344
0.1673	0.64	200	0.0538	0.0014	0.0014	0.4465	-0.0001	-365.8175	-394.9525	-0.7452	-0.8413
0.1395	0.96	300	0.0538	0.0016	0.0018	0.4410	-0.0002	-365.4079	-394.7079	-0.7549	-0.8511

Framework versions

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