metadata

library_name: transformers
license: gemma
base_model: google/gemma-7b
tags:
  - trl
  - orpo
  - generated_from_trainer
model-index:
  - name: gemma-7b-orpo-low-quality
    results: []

gemma-7b-orpo-low-quality

This model is a fine-tuned version of google/gemma-7b on the None dataset. It achieves the following results on the evaluation set:

Loss: 1.5398
Rewards/chosen: -0.0540
Rewards/rejected: -0.0625
Rewards/accuracies: 0.5396
Rewards/margins: 0.0085
Logps/rejected: -1.2503
Logps/chosen: -1.0803
Logits/rejected: 271.8756
Logits/chosen: 300.6891
Nll Loss: 1.4724
Log Odds Ratio: -0.6945
Log Odds Chosen: 0.2937

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
1.441	0.9955	167	1.4762	-0.0510	-0.0574	0.5324	0.0064	-1.1485	-1.0204	290.1581	318.9965	1.4310	-0.6990	0.1934
1.0908	1.9970	335	1.4250	-0.0497	-0.0576	0.5324	0.0079	-1.1528	-0.9950	285.8206	314.6779	1.3697	-0.6970	0.2360
0.5724	2.9866	501	1.5398	-0.0540	-0.0625	0.5396	0.0085	-1.2503	-1.0803	271.8756	300.6891	1.4724	-0.6945	0.2937

Framework versions

Transformers 4.44.2
Pytorch 2.4.0+cu121
Datasets 3.0.0
Tokenizers 0.19.1