abcdabcd987/viggo-llama2-7b-lora-16

Punica

Punica: Serving multiple LoRA finetuned LLMs at the cost of one

Paper: https://arxiv.org/abs/2310.18547

See https://github.com/punica-ai/punica/tree/master/examples/finetune

Model

• Base Model: Llama-2-7b-hf
• LoRA target: q_proj,k_proj,v_proj,o_proj,gate_proj,up_proj,down_proj
• LoRA rank: 16

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 4.0

Framework versions

• Transformers 4.34.1
• Pytorch 2.2.0.dev20230911+cu121
• Datasets 2.14.4
• Tokenizers 0.14.1