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

This model is a fine-tuned version of DUAL-GPO/phi-2-gpo-test-longest-iter-random2-1 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.0018
• Rewards/chosen: -0.0060
• Rewards/rejected: -0.0057
• Rewards/accuracies: 0.4955
• Rewards/margins: -0.0003
• Logps/rejected: -279.5291
• Logps/chosen: -307.4422
• Logits/rejected: 0.0341
• Logits/chosen: -0.0650

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	0.0017	-0.0031	-0.0027	0.4900	-0.0004	-279.2278	-307.1495	0.0468	-0.0515
0.0011	3.2	200	0.0019	-0.0041	-0.0041	0.5005	0.0000	-279.3705	-307.2468	0.0071	-0.0933

Framework versions

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