amielle's picture
chore: Remove redundant note in variable label
a18ded0
raw history blame
No virus
3.63 kB
import gradio as gr
from util import summarizer, textproc
from util.examples import entries
import importlib
importlib.reload(summarizer)
importlib.reload(textproc)
model_collection = summarizer.init_models()
patent_summarizer = summarizer.PatentSummarizer(model_collection)
iface = gr.Interface(
patent_summarizer.pipeline,
theme="huggingface",
examples=entries,
examples_per_page=4,
inputs=[
gr.inputs.Textbox(label="Patent Information",
placeholder="US10125002B2 or https://patents.google.com/patent/US10125002B2 ...",
lines=1,
default="US10125002B2"),
gr.inputs.CheckboxGroup(summarizer.summary_options,
default=summarizer.summary_options,
label="Summaries to Generate"),
gr.inputs.Dropdown(summarizer.model_names,
default=summarizer.model_names[3],
label="Abstract Model"),
gr.inputs.Dropdown(summarizer.model_names,
default=summarizer.model_names[0],
label="Background Model"),
gr.inputs.Dropdown(summarizer.model_names,
default=summarizer.model_names[2],
label="Claims Model"),
gr.inputs.Checkbox(default=True,
label="Collate Claims",
optional=False),
gr.inputs.Slider(minimum=250,
maximum=1000,
step=10,
default=250,
label="Input Document Word Limit"),
],
outputs=[
gr.outputs.Textbox(label="Abstract Summary"),
gr.outputs.Textbox(label="Background Summary"),
gr.outputs.Textbox(label="Sample Claims Summary")
],
title="Patent Summarizer πŸ“–",
description="""
v.1.0.0
Reading through patent documents is oftentimes a long and tedious task.
There are cases wherein one has to manually go through several pages in
order to determine if the patent is relevant to the prior art search.
This application is meant to automate the initial phase of going through
patents so that the potential inventor or researcher may lessen the time
spent trying to filter documents.
The Patent Summarizer:
✏️ Provides an interface for user input
πŸ“‚ Retrieves and parses the document from Patents Google based on the given patent information
πŸ“‘ Returns summaries for the abstract, background, and/or claims.
Notes:
- Increasing 'Input Document Word Limit' may improve results but will
cause inference time to increase
- Setting 'Collate Claims' to False will apply summarization across
individual claims. Doing so will greatly increase inference time.
Models explored:
πŸ€– Big Bird: Transformers for Longer Sequences
https://arxiv.org/abs/2007.14062 , https://huggingface.co/google/bigbird-pegasus-large-bigpatent
πŸ€– T5
https://arxiv.org/pdf/1910.10683.pdf , https://huggingface.co/cnicu/t5-small-booksum
πŸ€– BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension
https://arxiv.org/abs/1910.13461, https://huggingface.co/sshleifer/distilbart-cnn-6-6
πŸ€– PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization
https://arxiv.org/abs/1912.08777 , https://huggingface.co/google/pegasus-xsum
""",
)
iface.launch(enable_queue=True)