phi-2-gpo-test-longest-iter-random2-4

This model is a fine-tuned version of DUAL-GPO/phi-2-gpo-test-longest-iter-random2-3 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.0019
• Rewards/chosen: -0.0074
• Rewards/rejected: -0.0063
• Rewards/accuracies: 0.4710
• Rewards/margins: -0.0012
• Logps/rejected: -279.6524
• Logps/chosen: -307.5768
• Logits/rejected: 0.0429
• Logits/chosen: -0.0563

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
0.001	1.6	100	0.0018	-0.0035	-0.0023	0.4785	-0.0012	-279.2534	-307.1775	0.0583	-0.0400
0.0009	3.2	200	0.0019	-0.0082	-0.0066	0.4565	-0.0015	-279.6910	-307.6504	0.0455	-0.0553

Framework versions

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