LongForm-T5-XL / README.md
akoksal's picture
Update README.md
02205bf
|
raw
history blame
5.42 kB
metadata
language:
  - en
  - multilingual
  - de
  - it
  - es
  - fr
tags:
  - instruction-tuning
  - text-generation-inference
  - text2text-generation
widget:
  - text: Write an essay about meditation.
    example_title: Essay Generation
  - text: Give me 5 steps to clean my room.
    example_title: How-to Instructions
  - text: How are the continents formed?
    example_title: Question-Answering
  - text: >-
      Prompt: A man draws a gun in a dark alley and asks for your wallet. You
      begrudgingly obey. He throws it on the ground, shoots it till it
      screeches, and turns to you; 'you are safe now'. Write a story about given
      prompt.
    example_title: Story Generation
  - text: >-
      Write directions of a cooking recipe with these ingredients: chicken
      breast, carrots, green peas, celery, butter, onion, flour, salt, black
      pepper, celery seed, chicken broth, milk, unbaked pie crusts
    example_title: Recipe Generation
  - text: Schreiben Sie einen Blogbeitrag über die Vorteile des Lesens von Büchern.
    example_title: German Essay Generation
inference:
  parameters:
    top_p: 0.9
    do_sample: true
    max_length: 75
datasets:
  - akoksal/LongForm

LongForm-T5-XL

The LongForm dataset is created by leveraging English corpus examples with augmented instructions. We select a diverse set of human-written documents from existing corpora such as C4 and Wikipedia and generate instructions for the given documents via LLMs. Then, we extend these examples with structured corpora examples such as Stack Exchange and WikiHow and task examples such as question answering, email writing, grammar error correction, story/poem generation, and text summarization.

Github Repo: https://github.com/akoksal/LongForm

How to Load

import torch
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
model = AutoModelForSeq2SeqLM.from_pretrained("akoksal/LongForm-T5-XL")
tokenizer = AutoTokenizer.from_pretrained("akoksal/LongForm-T5-XL")

instruction = "Write an essay about meditation."
torch.manual_seed(42)
input_ids = tokenizer(instruction, return_tensors="pt").input_ids
target_ids = model.generate(input_ids, do_sample=True, max_new_tokens=50, top_p=0.9)
tokenizer.decode(target_ids[0], skip_special_tokens=True)
# Output:
# > Meditation is an ancient, spiritual practice. Meditation was first\
# practiced as early as 3000 BC by Indians. Meditation has been practiced\
# by people for thousands of years. People meditate in order to become more\
# present in their life. Meditation is

Evaluation

We provide in-depth evaluation of LongForm models and baselines in the paper. We present the METEOR scores of models in out-of-domain datasets. In all tasks, Recipe Generation (RGen), long-form question answering (ELI5), short story generation (WritingPrompts/WP), LongForm models outperform prior instruction-tuned models.

All Recipe Generation ELI5 Writing Prompts
T0++ 10.9 18.7 3.8 10.2
Tk-Instruct 6.3 12.9* 3.6 2.4
Flan-T5 10.6 20.9* 3.5 7.4
Alpaca-LLaMA-7B 14.6 19.5 12.5 11.8
OPT-30B 11.1 18.6 12.2 2.6
LongForm-T5-XL 16.3 20.2 18.3 10.6
LongForm-OPT-2.7B 17.8 15.5 17.9 19.9
LongForm-OPT-6.7B 17.7 16.9 17.2 19.0
LongForm-LLaMA-7B 19.7 21.7 18.6 18.9

Smaller versions of LongForm-OPT models are also available:

‡: We can just release the difference between LongForm-LLaMA-7B and pretrained LLaMA-7B publicly due to restrictions of LLaMA models.

Limitations

The LongForm dataset and models mainly focus on long text generation and have limitations regarding structured prediction tasks in NLP. Additionally, we observe that LongForm models may present hallucination problems similar to those found in LLMs.

License

The LongForm project is subject to a MIT License with custom limitations for restrictions imposed by OpenAI (for the instruction generation part), as well as the license of language models (OPT, LLaMA, and T5).

Citation

@misc{koksal2023longform,
      title={LongForm: Optimizing Instruction Tuning for Long Text Generation with Corpus Extraction}, 
      author={Abdullatif Köksal and Timo Schick and Anna Korhonen and Hinrich Schütze},
      year={2023},
      eprint={2304.08460},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}