Hugging Face's logo

DUAL-GPO
/
phi-2-gpo-iter-0

PEFT
Safetensors
phi
alignment-handbook
Generated from Trainer
trl
dpo
custom_code
Model card Files Community
phi-2-gpo-iter-0 / README.md
lole25's picture
lole25
End of training
21a61b1 verified
preview code
|
raw history blame contribute delete
No virus
1.65 kB
metadata
license: mit
library_name: peft
tags:
  - alignment-handbook
  - generated_from_trainer
  - trl
  - dpo
  - generated_from_trainer
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
base_model: microsoft/phi-2
model-index:
  - name: phi-2-gpo-iter-0
    results: []

phi-2-gpo-iter-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: 0.0103
  • Rewards/chosen: 0.0006
  • Rewards/rejected: 0.0005
  • Rewards/accuracies: 0.4945
  • Rewards/margins: 0.0001
  • Logps/rejected: -233.4960
  • Logps/chosen: -256.5553
  • Logits/rejected: 0.8946
  • Logits/chosen: 0.8373

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

Framework versions

  • PEFT 0.7.1
  • Transformers 4.36.2
  • Pytorch 2.1.2+cu118
  • Datasets 2.14.6
  • Tokenizers 0.15.2