phi-2-gpo-test-longest-iter-v1-2

This model is a fine-tuned version of DUAL-GPO/phi-2-gpo-test-longest-iter-v1-1 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.0016
• Rewards/chosen: -0.0008
• Rewards/rejected: -0.0008
• Rewards/accuracies: 0.4910
• Rewards/margins: 0.0001
• Logps/rejected: -278.6518
• Logps/chosen: -306.3463
• Logits/rejected: 0.0888
• Logits/chosen: -0.0087

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: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0011	1.6	100	0.0016	-0.0007	-0.0008	0.4960	0.0001	-278.6544	-306.3417	0.0909	-0.0073
0.0011	3.2	200	0.0016	-0.0001	-0.0005	0.4925	0.0003	-278.6177	-306.2834	0.0921	-0.0047

Framework versions

• PEFT 0.7.1
• Transformers 4.36.2
• Pytorch 2.2.1+cu121
• Datasets 2.14.6
• Tokenizers 0.15.2