phi-2-gpo-test-longest-iter-v1-0

This model is a fine-tuned version of lole25/phi-2-sft-ultrachat-lora on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.0004
• Rewards/chosen: 0.0012
• Rewards/rejected: 0.0010
• Rewards/accuracies: 0.4995
• Rewards/margins: 0.0002
• Logps/rejected: -233.4380
• Logps/chosen: -256.4973
• Logits/rejected: 0.8990
• Logits/chosen: 0.8417

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
0.0003	1.6	100	0.0004	0.0006	0.0004	0.4855	0.0002	-233.5017	-256.5565	0.8960	0.8387
0.0003	3.2	200	0.0004	0.0013	0.0009	0.5100	0.0004	-233.4492	-256.4811	0.8984	0.8412

Framework versions

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