cygu/llama-2-7b-logit-watermark-distill-kgw-k1-gamma0.25-delta1

Model description

Logit-based watermark distilled Llama 2 7B using the KGW $k=1,\gamma =0.25,\delta =1k=1,\; \backslash gamma=0.25,\; \backslash delta=1$ watermarking strategy in the paper On the Learnability of Watermarks for Language Models.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• training_steps: 5000

Framework versions

• Transformers 4.29.2
• Pytorch 2.0.1+cu117
• Datasets 2.13.1
• Tokenizers 0.13.3