Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training
Paper
• 2407.09121 • Published
• 6
Meta-Llama-3-8B-Instruct_derta_100step
This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the Evol-Instruct and BeaverTails dataset. The model is continued to train 100 steps with DeRTa on LLaMA3-8B-Instruct.
Model description
Please refer to the paper Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training and GitHub DeRTa.
Input format:
[INST] Your Instruction [\INST]
Intended uses & limitations
The model is trained with DeRTa, showing a high safety performance.
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- weight_decay: 2e-5
- eval_batch_size: 1
- seed: 1
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 128
- total_eval_batch_size: 8
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.03
- num_epochs: 2.0
Training results
Framework versions
- Transformers 4.40.0
- Pytorch 2.2.0+cu118
- Datasets 2.10.0
- Tokenizers 0.19.1
- Downloads last month
- 5
Model tree for Youliang/llama3-8b-instruct-derta-100step
Base model
meta-llama/Meta-Llama-3-8B-Instruct