Lsg Bart Base 4096 Pubmed

🚀 ccdv/lsg-bart-base-4096-pubmed

This model is a fine - tuned version of ccdv/lsg-bart-base-4096 on the scientific_papers pubmed dataset. It is designed for text summarization tasks and achieves remarkable results on the test set.

🚀 Quick Start

Transformers >= 4.36.1 This model relies on a custom modeling file, you need to add trust_remote_code=True See #13467

LSG ArXiv paper. Github/conversion script is available at this link.

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline

tokenizer = AutoTokenizer.from_pretrained("ccdv/lsg-bart-base-4096-pubmed", trust_remote_code=True)
model = AutoModelForSeq2SeqLM.from_pretrained("ccdv/lsg-bart-base-4096-pubmed", trust_remote_code=True)

text = "Replace by what you want."
pipe = pipeline("text2text-generation", model=model, tokenizer=tokenizer, device=0)
generated_text = pipe(
  text, 
  truncation=True, 
  max_length=64, 
  no_repeat_ngram_size=7,
  num_beams=2,
  early_stopping=True
  )

✨ Features

• Long Sequence Handling: The model relies on Local - Sparse - Global attention to handle long sequences, as shown in the figure below:
• Fine - Tuned Performance: It is fine - tuned on the scientific_papers pubmed dataset, achieving good results on multiple metrics such as ROUGE.

📚 Documentation

Test Set Results

It achieves the following results on the test set:

With smaller block size (lower ressources):

Model Details

The model has about ~145 millions parameters (6 encoder layers - 6 decoder layers). It is warm - started from BART - base, converted to handle long sequences (encoder only) and fine - tuned.

Training and Evaluation

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 8e-05
• train_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 8.0

Generate hyperparameters

The following hyperparameters were used during generation:

• dataset_name: scientific_papers
• dataset_config_name: pubmed
• eval_batch_size: 8
• eval_samples: 6658
• early_stopping: True
• ignore_pad_token_for_loss: True
• length_penalty: 2.0
• max_length: 512
• min_length: 128
• num_beams: 5
• no_repeat_ngram_size: None
• seed: 123

Framework versions

• Transformers 4.18.0
• Pytorch 1.10.1+cu102
• Datasets 2.1.0
• Tokenizers 0.11.6

🔧 Technical Details

The model leverages Local - Sparse - Global attention mechanism to handle long - sequence inputs effectively. This attention mechanism allows the model to focus on relevant parts of the long text while reducing computational complexity. The model architecture consists of 6 encoder layers and 6 decoder layers, with approximately 145 million parameters. It starts from the BART - base model, is modified to handle long sequences (only in the encoder part), and then is fine - tuned on the scientific_papers pubmed dataset.