Longformer Encoder-Decoder (LED) fine-tuned on Booksum

• This model is a fine-tuned version of allenai/led-large-16384 on the booksum dataset.
• the goal was to create a model that can generalize well and is useful in summarizing lots of text in academic and daily usage.
• I don't want to read anymore give me a notebook implementation
• all the parameters for generation on the API are the same as the base model for easy comparison between versions.
• works well on lots of text, can hand 16384 tokens/batch.

Note: the API is set to generate a max of 64 tokens for runtime reasons, so the summaries may be truncated (depending on length of input text). For best results use python as below.

Usage - Basics

• it is recommended to use encoder_no_repeat_ngram_size=3 when calling the pipeline object to improve summary quality.
  • this param forces the model to use new vocabulary and create an abstractive summary, otherwise it may compile the best extractive summary from the input provided.
• create the pipeline object:

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
from transformers import pipeline

hf_name = 'pszemraj/led-large-book-summary'

_model = AutoModelForSeq2SeqLM.from_pretrained(
                hf_name,
                low_cpu_mem_usage=True,
            )

_tokenizer = AutoTokenizer.from_pretrained(
                hf_name
            )
                                           

summarizer = pipeline(
                    "summarization", 
                    model=_model, 
                    tokenizer=_tokenizer
                )

• put words into the pipeline object:

wall_of_text = "your words here"

result = summarizer(
           wall_of_text,
           min_length=16, 
           max_length=256,
           no_repeat_ngram_size=3, 
           encoder_no_repeat_ngram_size =3,
           clean_up_tokenization_spaces=True,
           repetition_penalty=3.7,
           num_beams=4,
           early_stopping=True,
    )

• Important: To generate the best quality summaries, you should use the global attention mask when decoding, as demonstrated in this community notebook here, see the definition of generate_answer(batch).
• If you run into compute constraints, try the base version pszemraj/led-base-book-summary

Training and evaluation data

• the booksum dataset
• During training, the input text was the text of the chapter, and the output was summary_text
• Eval results can be found here with metrics on the sidebar.

Training procedure

• Training completed on the BookSum dataset for 13 total epochs
• The final four epochs combined the training and validation sets as 'train' in an effort to increase generalization.

Training hyperparameters

Initial Three Epochs

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

In-between Epochs

Unfortunately, don't have all records on-hand for middle epochs, the following should be representative:

• learning_rate: 4e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 6 (in addition to prior model)

Final Two Epochs

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 2 (in addition to prior model)

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1