metadata

library_name: transformers
license: gemma
base_model: google/gemma-7b
tags:
  - alignment-handbook
  - trl
  - orpo
  - generated_from_trainer
  - trl
  - orpo
  - generated_from_trainer
datasets:
  - silviasapora/low_quality_dpo7k
model-index:
  - name: gemma-7b-borpo-low-quality-v2
    results: []

gemma-7b-borpo-low-quality-v2

This model is a fine-tuned version of google/gemma-7b on the silviasapora/low_quality_dpo7k dataset. It achieves the following results on the evaluation set:

Loss: 1.6017
Rewards/chosen: -0.0578
Rewards/rejected: -0.0690
Rewards/accuracies: 0.5714
Rewards/margins: 0.0112
Logps/rejected: -1.3795
Logps/chosen: -1.1561
Logits/rejected: 249.0934
Logits/chosen: 304.2649
Nll Loss: 1.5643
Log Odds Ratio: -0.6745
Log Odds Chosen: 0.3316

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: 2
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 2
total_train_batch_size: 32
total_eval_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: inverse_sqrt
lr_scheduler_warmup_steps: 100
num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Nll Loss	Log Odds Ratio	Log Odds Chosen
1.4218	1.0	168	1.4488	-0.0504	-0.0580	0.5571	0.0076	-1.1591	-1.0071	273.7526	326.8029	1.4553	-0.6712	0.2324
1.0804	2.0	336	1.4225	-0.0511	-0.0591	0.5143	0.0080	-1.1830	-1.0220	278.2473	330.5067	1.4083	-0.6897	0.2152
0.5651	3.0	504	1.6017	-0.0578	-0.0690	0.5714	0.0112	-1.3795	-1.1561	249.0934	304.2649	1.5643	-0.6745	0.3316

Framework versions

Transformers 4.44.2
Pytorch 2.4.1+cu121
Datasets 3.0.0
Tokenizers 0.19.1