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
    results: []

gemma-7b-borpo-low-quality

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: 1.5380
Rewards/chosen: -0.0547
Rewards/rejected: -0.0625
Rewards/accuracies: 0.5468
Rewards/margins: 0.0079
Logps/rejected: -1.2508
Logps/chosen: -1.0933
Logits/rejected: 267.2346
Logits/chosen: 296.6808
Nll Loss: 1.4703
Log Odds Ratio: -0.7039
Log Odds Chosen: 0.2721

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: 4
gradient_accumulation_steps: 4
total_train_batch_size: 32
total_eval_batch_size: 4
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
1.436	0.9955	167	1.4639	-0.0502	-0.0571	0.5540	0.0068	-1.1413	-1.0048	294.2689	322.9157	1.4152	-0.6882	0.2192
1.0918	1.9970	335	1.4233	-0.0501	-0.0574	0.4964	0.0073	-1.1475	-1.0012	284.8744	313.3100	1.3661	-0.7028	0.2209
0.576	2.9866	501	1.5380	-0.0547	-0.0625	0.5468	0.0079	-1.2508	-1.0933	267.2346	296.6808	1.4703	-0.7039	0.2721

Framework versions

Transformers 4.44.2
Pytorch 2.4.0+cu121
Datasets 3.0.0
Tokenizers 0.19.1