aya-advprompter

This model is a fine-tuned version of CohereForAI/aya-23-8B on the simonycl/aya-23-8B_advprompter_jailbreak dataset. It achieves the following results on the evaluation set:

• Loss: 0.0459
• Rewards/chosen: 0.0182
• Rewards/rejected: -6.7884
• Rewards/accuracies: 1.0
• Rewards/margins: 6.8065
• Logps/rejected: -867.2261
• Logps/chosen: -114.6688
• Logits/rejected: 0.0796
• Logits/chosen: -0.2307

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-07
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• total_eval_batch_size: 2
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Logits/chosen	Logits/rejected	Logps/chosen	Logps/rejected	Validation Loss	Rewards/accuracies	Rewards/chosen	Rewards/margins	Rewards/rejected
0.5229	0.3612	30	-0.4619	-0.3434	-98.2886	-212.0101	0.5059	1.0	0.1820	0.4182	-0.2362
0.2411	0.7223	60	-0.4067	-0.2327	-88.9001	-330.7860	0.2135	1.0	0.2758	1.6998	-1.4240
0.0634	1.0835	90	-0.2580	-0.0357	-99.5121	-607.3592	0.0751	1.0	0.1697	4.3594	-4.1897
0.0452	1.4454	120	0.0532	0.0757	-5.9396	1.0	6.0153	-782.3494	-108.9159	0.0380	-0.2345
0.0307	1.8066	150	0.0459	0.0182	-6.7884	1.0	6.8065	-867.2261	-114.6688	0.0796	-0.2307

Framework versions

• Transformers 4.44.2
• Pytorch 2.3.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1