gemma-7b-borpo-low-quality-v3

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: 2.1095
• Rewards/chosen: -0.6954
• Rewards/rejected: -0.8346
• Rewards/accuracies: 0.5571
• Rewards/margins: 0.1392
• Logps/rejected: -1.6692
• Logps/chosen: -1.3909
• Logits/rejected: 262.5518
• Logits/chosen: 319.3429
• Nll Loss: 1.7836
• Log Odds Ratio: -0.6395
• Log Odds Chosen: 0.4455

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: 1e-05
• 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
1.9721	1.0	168	1.9526	-0.6072	-0.7027	0.5571	0.0955	-1.4054	-1.2144	282.1215	336.2867	1.6515	-0.6573	0.2649
1.3299	2.0	336	1.9015	-0.5986	-0.6805	0.5	0.0820	-1.3611	-1.1972	293.2820	345.2333	1.5933	-0.6792	0.2173
0.6266	3.0	504	2.1095	-0.6954	-0.8346	0.5571	0.1392	-1.6692	-1.3909	262.5518	319.3429	1.7836	-0.6395	0.4455

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1