README.md

---
base_model: meta-llama/Llama-2-7b-hf
tags:
- generated_from_trainer
model-index:
- name: Llama2-7bn-xsum-adapter
  results: []
datasets:
- EdinburghNLP/xsum
language:
- en
pipeline_tag: summarization
metrics:
- rouge
---

<!-- 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-adapter

Weights & Biases runs for training and evaluation are available for a detailed overview!

This model is a fine-tuned version of [meta-llama/Llama-2-7b-hf](https://huggingface.co/meta-llama/Llama-2-7b-hf) 
on a [XSum](https://huggingface.co/datasets/EdinburghNLP/xsum) dataset with Causal LM task. You can view all the implementation details on the [GitHub project](https://github.com/ernlavr/llamarizer)


## Weights & Biases Training and Evaluation Documentation

See the [training](https://wandb.ai/ernlavr/adv_nlp2023/runs/yk6ytvv2) and 
[evaluation](https://wandb.ai/ernlavr/adv_nlp2023/runs/f41oo2c6?workspace=user-ernestslavrinovics)
on Weights & Biases for more details!

Summary table of final metrics:
| Metric               | rouge1  | rouge2  | rougeL  | FactCC  | ANLI   | SummaC  | BARTScore  |
|------------------------|---------|---------|---------|---------|--------|---------|------------|
| Mean                   | 0.18    | 0.033   | 0.126   | 0.188   | 0.408  | 0.658   | -3.713     |
| Std                    | 0.09    | 0.049   | 0.067   | 0.317   | 0.462  | 0.247   | 0.831      |



## Training procedure
Causal language modeling. Nesting the summary paragraph in a prompt: {Summarize this article: '<INPUT_DOCUMENT>'; Summary: <OUTPUT>}

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.001
- train_batch_size: 32
- eval_batch_size: 32
- 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: 450.5
- num_epochs: 3
- mixed_precision_training: Native AMP

### Framework versions

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