metadata

library_name: transformers
license: gemma
base_model: google/gemma-7b
tags:
  - alignment-handbook
  - trl
  - orpo
  - generated_from_trainer
  - trl
  - orpo
  - alignment-handbook
  - generated_from_trainer
datasets:
  - silviasapora/low_quality_dpo7k
model-index:
  - name: gemma-7b-borpo-low-quality-v4
    results: []

gemma-7b-borpo-low-quality-v4

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: 1.8577
Rewards/chosen: -0.5993
Rewards/rejected: -0.7602
Rewards/accuracies: 0.6143
Rewards/margins: 0.1610
Logps/rejected: -1.5205
Logps/chosen: -1.1986
Logits/rejected: 240.3907
Logits/chosen: 301.1215
Nll Loss: 1.5532
Log Odds Ratio: -0.6421
Log Odds Chosen: 0.4396

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: 4
total_train_batch_size: 64
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
1.8227	1.0	84	1.9616	-0.6050	-0.6743	0.5	0.0693	-1.3486	-1.2099	257.8447	315.1940	1.6719	-0.6903	0.1646
1.4803	2.0	168	1.7681	-0.5462	-0.6508	0.5286	0.1046	-1.3017	-1.0924	274.3526	328.0207	1.4854	-0.6718	0.2561
0.9109	3.0	252	1.8577	-0.5993	-0.7602	0.6143	0.1610	-1.5205	-1.1986	240.3907	301.1215	1.5532	-0.6421	0.4396

Framework versions

Transformers 4.44.2
Pytorch 2.4.1+cu121
Datasets 3.0.0
Tokenizers 0.19.1