Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Peter
commited on
Commit
·
fe0e9af
1
Parent(s):
4a607b7
:sparkles: add app mwe
Browse files
app.py
ADDED
|
@@ -0,0 +1,153 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import logging
|
| 2 |
+
from pathlib import Path
|
| 3 |
+
import os
|
| 4 |
+
import gradio as gr
|
| 5 |
+
import nltk
|
| 6 |
+
import torch
|
| 7 |
+
from cleantext import clean
|
| 8 |
+
|
| 9 |
+
from summarize import load_model_and_tokenizer, summarize_via_tokenbatches
|
| 10 |
+
|
| 11 |
+
_here = Path(__file__).parent
|
| 12 |
+
|
| 13 |
+
nltk.download("stopwords") # TODO=find where this requirement originates from
|
| 14 |
+
|
| 15 |
+
import transformers
|
| 16 |
+
|
| 17 |
+
transformers.logging.set_verbosity_error()
|
| 18 |
+
logging.basicConfig()
|
| 19 |
+
|
| 20 |
+
def truncate_word_count(text, max_words=512):
|
| 21 |
+
"""
|
| 22 |
+
truncate_word_count - a helper function for the gradio module
|
| 23 |
+
Parameters
|
| 24 |
+
----------
|
| 25 |
+
text : str, required, the text to be processed
|
| 26 |
+
max_words : int, optional, the maximum number of words, default=512
|
| 27 |
+
Returns
|
| 28 |
+
-------
|
| 29 |
+
dict, the text and whether it was truncated
|
| 30 |
+
"""
|
| 31 |
+
words = text.split()
|
| 32 |
+
processed = {}
|
| 33 |
+
if len(words) > max_words:
|
| 34 |
+
processed["was_truncated"] = True
|
| 35 |
+
processed["truncated_text"] = " ".join(words[:max_words])
|
| 36 |
+
else:
|
| 37 |
+
processed["was_truncated"] = False
|
| 38 |
+
processed["truncated_text"] = text
|
| 39 |
+
return processed
|
| 40 |
+
|
| 41 |
+
def proc_submission(
|
| 42 |
+
input_text: str,
|
| 43 |
+
num_beams,
|
| 44 |
+
length_penalty,
|
| 45 |
+
repetition_penalty,
|
| 46 |
+
no_repeat_ngram_size,
|
| 47 |
+
token_batch_length,
|
| 48 |
+
max_input_length: int = 512,
|
| 49 |
+
):
|
| 50 |
+
"""
|
| 51 |
+
proc_submission - a helper function for the gradio module
|
| 52 |
+
Parameters
|
| 53 |
+
----------
|
| 54 |
+
input_text : str, required, the text to be processed
|
| 55 |
+
max_input_length : int, optional, the maximum length of the input text, default=512
|
| 56 |
+
Returns
|
| 57 |
+
-------
|
| 58 |
+
str of HTML, the interactive HTML form for the model
|
| 59 |
+
"""
|
| 60 |
+
|
| 61 |
+
settings = {
|
| 62 |
+
"length_penalty": length_penalty,
|
| 63 |
+
"repetition_penalty": repetition_penalty,
|
| 64 |
+
"no_repeat_ngram_size": no_repeat_ngram_size,
|
| 65 |
+
"encoder_no_repeat_ngram_size": 4,
|
| 66 |
+
"num_beams": num_beams,
|
| 67 |
+
}
|
| 68 |
+
|
| 69 |
+
history = {}
|
| 70 |
+
clean_text = clean(input_text, lower=False)
|
| 71 |
+
processed = truncate_word_count(clean_text, max_input_length)
|
| 72 |
+
if processed["was_truncated"]:
|
| 73 |
+
history["input_text"] = processed["truncated_text"]
|
| 74 |
+
history["was_truncated"] = True
|
| 75 |
+
msg = f"Input text was truncated to {max_input_length} characters."
|
| 76 |
+
logging.warning(msg)
|
| 77 |
+
history["WARNING"] = msg
|
| 78 |
+
else:
|
| 79 |
+
history["input_text"] = input_text
|
| 80 |
+
history["was_truncated"] = False
|
| 81 |
+
|
| 82 |
+
_summaries = summarize_via_tokenbatches(
|
| 83 |
+
history["input_text"],
|
| 84 |
+
model, tokenizer,
|
| 85 |
+
batch_length=token_batch_length,
|
| 86 |
+
**settings,
|
| 87 |
+
)
|
| 88 |
+
sum_text = [s['summary'][0] for s in _summaries]
|
| 89 |
+
sum_scores = [f"\n - {round(s['summary_score'],4)}" for s in _summaries]
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
history["Input"] = input_text
|
| 93 |
+
history["Summary Text"] = "\n\t".join(sum_text)
|
| 94 |
+
history["Summary Scores"] = "\n".join(sum_scores)
|
| 95 |
+
html = ""
|
| 96 |
+
for name, item in history.items():
|
| 97 |
+
html += (
|
| 98 |
+
f"<h2>{name}:</h2><hr><b>{item}</b><br><br>"
|
| 99 |
+
if "summary" not in name.lower()
|
| 100 |
+
else f"<h2>{name}:</h2><hr><b>{item}</b>"
|
| 101 |
+
)
|
| 102 |
+
|
| 103 |
+
html += ""
|
| 104 |
+
|
| 105 |
+
return html
|
| 106 |
+
|
| 107 |
+
def load_examples(examples_dir='examples'):
|
| 108 |
+
src = _here / examples_dir
|
| 109 |
+
src.mkdir(exist_ok=True)
|
| 110 |
+
examples = [f for f in src.glob("*.txt")]
|
| 111 |
+
# load the examples into a list
|
| 112 |
+
text_examples = []
|
| 113 |
+
for example in examples:
|
| 114 |
+
with open(example, "r") as f:
|
| 115 |
+
text = f.read()
|
| 116 |
+
text_examples.append([text, 4, 2048, 0.7,3.5,3])
|
| 117 |
+
|
| 118 |
+
return text_examples
|
| 119 |
+
|
| 120 |
+
if __name__ == "__main__":
|
| 121 |
+
|
| 122 |
+
model, tokenizer = load_model_and_tokenizer('pszemraj/led-large-book-summary')
|
| 123 |
+
title = "Long-form text summarization with LED on the BookSumm dataset"
|
| 124 |
+
description = "This is a simple example of using the LED model to summarize a long-form text."
|
| 125 |
+
|
| 126 |
+
gr.Interface(
|
| 127 |
+
proc_submission,
|
| 128 |
+
inputs=[
|
| 129 |
+
gr.inputs.Textbox(lines=10, label="input text"),
|
| 130 |
+
gr.inputs.Slider(
|
| 131 |
+
minimum=4, maximum=10, label="num_beams", default=4, step=1
|
| 132 |
+
),
|
| 133 |
+
gr.Dropdown(choices=[512, 1024, 2048, 4096], label="token_batch_length", default=2048),
|
| 134 |
+
gr.inputs.Slider(
|
| 135 |
+
minimum=0.5, maximum=1.1, label="length_penalty", default=0.7, step=0.05
|
| 136 |
+
),
|
| 137 |
+
gr.inputs.Slider(
|
| 138 |
+
minimum=1.0,
|
| 139 |
+
maximum=5.0,
|
| 140 |
+
label="repetition_penalty",
|
| 141 |
+
default=3.5,
|
| 142 |
+
step=0.1,
|
| 143 |
+
),
|
| 144 |
+
gr.inputs.Slider(
|
| 145 |
+
minimum=2, maximum=4, label="no_repeat_ngram_size", default=3, step=1
|
| 146 |
+
),
|
| 147 |
+
],
|
| 148 |
+
outputs="html",
|
| 149 |
+
examples_per_page=4,
|
| 150 |
+
title=title,
|
| 151 |
+
description=description,
|
| 152 |
+
examples=load_examples(),
|
| 153 |
+
).launch(enable_queue=True, share=True)
|