metadata

license: mit
library_name: peft
tags:
  - alignment-handbook
  - generated_from_trainer
  - trl
  - dpo
  - generated_from_trainer
base_model: microsoft/phi-2
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
model-index:
  - name: phi-2-gpo-test-longest-iter-1
    results: []

phi-2-gpo-test-longest-iter-1

This model is a fine-tuned version of DUAL-GPO/phi-2-gpo-test-longest-iter-0 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

Loss: 0.0108
Rewards/chosen: -0.0007
Rewards/rejected: -0.0005
Rewards/accuracies: 0.4995
Rewards/margins: -0.0002
Logps/rejected: -278.7133
Logps/chosen: -306.4168
Logits/rejected: 0.0867
Logits/chosen: -0.0107

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: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0103	1.6	100	0.0108	-0.0015	-0.0014	0.5010	-0.0001	-278.8064	-306.4953	0.0771	-0.0200

Framework versions

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