llama2-7bn-xsum-cnn-lora-adapter

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

llama2-7bn-xsum-cnn-adapter

This model is a fine-tuned version of meta-llama/Llama-2-7b-hf on XSum and CNN/DM. LoRA adapter model based on LLama2 7bn. You can view all the implementation details on the GitHub project

Weights and Biases Documentation: Training and Eval

See Weights and Biases for training details.

Training procedure

• Input source document wrapped in a prompt: "Summarize the following article:<INPUT>; Summary: <OUTPUT>"
• Trained using cross-entropy on CausalLM task
• Data splits consist of sequences up to 512 tokens:
  • Training n-datapoints: 115'354 XSum; 27494 CNN
  • Val n-datapoints: 3928 XSum; 1211 CNN

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• lr_scheduler_warmup_steps: 558.0
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Achieves loss=2.021 on valdiation split, see W&B run (link above) for more details.

Framework versions

• Transformers 4.35.0
• Pytorch 2.0.1
• Datasets 2.14.6
• Tokenizers 0.14.1