# coding=utf-8
# Copyright 2023 The GlotLID Authors.
# Lint as: python3
# This space is built based on AMR-KELEG/ALDi space.
# GlotLID Space
import string
import constants
import pandas as pd
import streamlit as st
from huggingface_hub import hf_hub_download
from GlotScript import get_script_predictor
import matplotlib
from matplotlib import pyplot as plt
import fasttext
import altair as alt
from altair import X, Y, Scale
import base64
import json
import os
import re
import transformers
from transformers import pipeline
@st.cache_resource
def load_sp():
sp = get_script_predictor()
return sp
sp = load_sp()
def get_script(text):
"""Get the writing systems of given text.
Args:
text: The text to be preprocessed.
Returns:
The main script and list of all scripts.
"""
res = sp(text)
main_script = res[0] if res[0] else 'Zyyy'
all_scripts_dict = res[2]['details']
if all_scripts_dict:
all_scripts = list(all_scripts_dict.keys())
else:
all_scripts = 'Zyyy'
for ws in all_scripts:
if ws in ['Kana', 'Hrkt', 'Hani', 'Hira']:
all_scripts.append('Jpan')
all_scripts = list(set(all_scripts))
return main_script, all_scripts
def preprocess_text(text):
"""Apply preprocessing to the given text.
Args:
text: Thetext to be preprocessed.
Returns:
The preprocessed text.
"""
# remove \n
text = text.replace('\n', ' ')
# get rid of characters that are ubiquitous
replace_by = " "
replacement_map = {
ord(c): replace_by
for c in ':•#{|}' + string.digits
}
text = text.translate(replacement_map)
# make multiple space one space
text = re.sub(r'\s+', ' ', text)
# strip the text
text = text.strip()
return text
@st.cache_data
def language_names(json_path):
with open(json_path, 'r') as json_file:
data = json.load(json_file)
return data
label2name = language_names("assets/language_names.json")
def get_name(label):
"""Get the name of language from label"""
iso_3 = label.split('_')[0]
name = label2name[iso_3]
return name
@st.cache_data
def render_svg(svg):
"""Renders the given svg string."""
b64 = base64.b64encode(svg.encode("utf-8")).decode("utf-8")
html = rf' 
'
c = st.container()
c.write(html, unsafe_allow_html=True)
@st.cache_data
def render_metadata():
"""Renders the metadata."""
html = r"""
"""
c = st.container()
c.write(html, unsafe_allow_html=True)
@st.cache_data
def citation():
"""Renders the metadata."""
_CITATION = """
@inproceedings{
kargaran2023glotlid,
title={GlotLID: Language Identification for Low-Resource Languages},
author={Kargaran, Amir Hossein and Imani, Ayyoob and Yvon, Fran{\c{c}}ois and Sch{\"u}tze, Hinrich},
booktitle={The 2023 Conference on Empirical Methods in Natural Language Processing},
year={2023},
url={https://openreview.net/forum?id=dl4e3EBz5j}
}"""
st.code(_CITATION, language="python", line_numbers=False)
@st.cache_data
def convert_df(df):
# IMPORTANT: Cache the conversion to prevent computation on every rerun
return df.to_csv(index=None).encode("utf-8")
@st.cache_resource
def load_model(model_name, file_name):
model_path = hf_hub_download(repo_id=model_name, filename=file_name)
model = fasttext.load_model(model_path)
return model
@st.cache_resource
def load_model_pipeline(model_name, file_name):
model = pipeline("text-classification", model=model_name)
return model
# model_1 = load_model(constants.MODEL_NAME, "model_v1.bin")
# model_2 = load_model(constants.MODEL_NAME, "model_v2.bin")
# model_3 = load_model(constants.MODEL_NAME, "model_v3.bin")
# openlid = load_model('laurievb/OpenLID', "model.bin")
# nllb = load_model('facebook/fasttext-language-identification', "model.bin")
# MODELS
model_xlmr_large = load_model_pipeline('dsfsi/za-xlmrlarge-lid', "model.bin")
model_serengeti = load_model_pipeline('dsfsi/za-serengeti-lid', "model.bin")
model_afriberta = load_model_pipeline('dsfsi/za-afriberta-lid', "model.bin")
model_afroxlmr_base = load_model_pipeline('dsfsi/za-afro-xlmr-base-lid', "model.bin")
model_afrolm = load_model_pipeline('dsfsi/za-afrolm-lid', "model.bin")
za_lid = load_model_pipeline('dsfsi/za-lid-bert', "model.bin")
openlid = load_model('laurievb/OpenLID', "model.bin")
glotlid_3 = load_model(constants.MODEL_NAME, "model_v3.bin")
# @st.cache_resource
def plot(label, prob):
ORANGE_COLOR = "#FF8000"
BLACK_COLOR = "#31333F"
fig, ax = plt.subplots(figsize=(8, 1))
fig.patch.set_facecolor("none")
ax.set_facecolor("none")
ax.spines["left"].set_color(BLACK_COLOR)
ax.spines["bottom"].set_color(BLACK_COLOR)
ax.tick_params(axis="x", colors=BLACK_COLOR)
ax.spines[["right", "top"]].set_visible(False)
ax.barh(y=[0], width=[prob], color=ORANGE_COLOR)
ax.set_xlim(0, 1)
ax.set_ylim(-1, 1)
ax.set_title(f"Label: {label}, Language: {get_name(label)}", color=BLACK_COLOR)
ax.get_yaxis().set_visible(False)
ax.set_xlabel("Confidence", color=BLACK_COLOR)
st.pyplot(fig)
# @st.cache_resource
def plot_multiples(models, labels, probs):
ORANGE_COLOR = "#FF8000"
BLACK_COLOR = "#31333F"
fig, ax = plt.subplots(figsize=(12, len(models)))
fig.patch.set_facecolor("none")
ax.set_facecolor("none")
ax.spines["left"].set_color(BLACK_COLOR)
ax.spines["bottom"].set_color(BLACK_COLOR)
ax.tick_params(axis="x", colors=BLACK_COLOR)
ax.spines[["right", "top"]].set_visible(False)
# Plot bars for each model, label, and probability
y_positions = range(len(models)) # Y positions for each model
ax.barh(y=y_positions, width=probs, color=ORANGE_COLOR)
# Add labels next to each bar
for i, (prob, label) in enumerate(zip(probs, labels)):
ax.text(prob + 0.01, i, f"{label} ({prob:.2f})", va='center', color=BLACK_COLOR)
# Set y-ticks and labels
ax.set_yticks(y_positions)
ax.set_yticklabels(models, color=BLACK_COLOR)
ax.set_xlim(0, 1)
ax.set_xlabel("Confidence", color=BLACK_COLOR)
ax.set_title("Model Predictions", color=BLACK_COLOR)
st.pyplot(fig)
def compute(sentences, version = 'v3'):
"""Computes the language probablities and labels for the given sentences.
Args:
sentences: A list of sentences.
Returns:
A list of language probablities and labels for the given sentences.
"""
progress_text = "Computing Language..."
if version == 'xlmrlarge':
model_choice = model_xlmr_large
elif version == 'serengeti':
model_choice = model_serengeti
elif version == 'afriberta':
model_choice = model_afriberta
elif version == 'afroxlmrbase':
model_choice = model_afroxlmr_base
elif version=='afrolm':
model_choice = model_afrolm
elif version == 'BERT':
model_choice = za_lid
elif version == 'openlid-201':
model_choice = openlid
elif version == 'GlotLID v3':
model_choice = glotlid_3
else:
model_choice = [(model_xlmr_large, "xlmrlarge"),(model_serengeti,"serengeti"), (model_afriberta,"afriberta"), (model_afroxlmr_base,"afroxlmrbase"), (model_afrolm,"afrolm"), (za_lid,"BERT"), (openlid,"openlid-201"), (glotlid_3,"GlotLID v3")]
my_bar = st.progress(0, text=progress_text)
probs = []
labels = []
sentences = [preprocess_text(sent) for sent in sentences]
for index, sent in enumerate(sentences):
if type(model_choice) == list:
all_models_pred = []
for model_version in model_choice:
m_version = model_version[1]
model = model_version[0]
if m_version not in ["openlid-201", "GlotLID v3"]:
output = model.predict(sent)
output_label = output[index]['label']
output_prob = output[index]['score']
output_label_language = output[index]['label']
labels = labels + [output_label]
probs = probs + [output_prob]
my_bar.progress(
min((index) / len(sentences), 1),
text=progress_text,
)
else:
output = model.predict(sent)
output_label = output[0][0].split('__')[-1].replace('_Hans', '_Hani').replace('_Hant', '_Hani')
output_prob = max(min(output[1][0], 1), 0)
output_label_language = output_label.split('_')[0]
# script control
if version in ['GlotLID v3', 'openlid-201', 'nllb-218'] and output_label_language!= 'zxx':
main_script, all_scripts = get_script(sent)
output_label_script = output_label.split('_')[1]
if output_label_script not in all_scripts:
output_label_script = main_script
output_label = f"und_{output_label_script}"
output_prob = 0
labels = labels + [output_label]
probs = probs + [output_prob]
my_bar.progress(
min((index) / len(sentences), 1),
text=progress_text,
)
else:
if version not in ["openlid-201", "GlotLID v3"]:
output = model_choice.predict(sent)
output_label = output[index]['label']
output_prob = output[index]['score']
output_label_language = output[index]['label']
labels = labels + [output_label]
probs = probs + [output_prob]
my_bar.progress(
min((index) / len(sentences), 1),
text=progress_text,
)
else:
output = model_choice.predict(sent)
output_label = output[0][0].split('__')[-1].replace('_Hans', '_Hani').replace('_Hant', '_Hani')
output_prob = max(min(output[1][0], 1), 0)
output_label_language = output_label.split('_')[0]
# script control
if version in ['GlotLID v3', 'openlid-201', 'nllb-218'] and output_label_language!= 'zxx':
main_script, all_scripts = get_script(sent)
output_label_script = output_label.split('_')[1]
if output_label_script not in all_scripts:
output_label_script = main_script
output_label = f"und_{output_label_script}"
output_prob = 0
labels = labels + [output_label]
probs = probs + [output_prob]
my_bar.progress(
min((index) / len(sentences), 1),
text=progress_text,
)
my_bar.empty()
return probs, labels
# st.markdown("[](https://huggingface.co/spaces/cis-lmu/glotlid-space?duplicate=true)")
# render_svg(open("assets/glotlid_logo.svg").read())
render_metadata()
img1, img2, img3 = st.columns(3)
with img2:
with st.container(border=False):
st.image("logo_transparent_small.png")
st.markdown("**DSFSI** Language Identification (LID) Inference Endpoint Created with **HuggingFace Spaces**.")
with st.expander("More information about the space"):
st.write('''
Authors: Alexandre Lacoste, Alexandra Luccioni, Victor Schmidt, Thomas Dandres
''')
tab1, tab2 = st.tabs(["Input a Sentence", "Upload a File"])
with tab1:
# choice = st.radio(
# "Set granularity level",
# ["default", "merge", "individual"],
# captions=["enable both macrolanguage and its varieties (default)", "merge macrolanguage and its varieties into one label", "remove macrolanguages - only shows individual langauges"],
# )
version = st.radio(
"Choose model",
["xlmrlarge", "serengeti", "afriberta", "afroxlmrbase", "afrolm", "BERT", "openlid-201", "GlotLID v3", "All-Models"],
captions=["za-XLMR-Large", "za-Serengeti", "za-AfriBERTa", "za-Afro-XLMR-BASE", "za-AfroLM", "za-BERT", "OpenLID", "GlotLID v3",'All-Models'],
index = 4,
key = 'version_tab1',
horizontal = True
)
sent = st.text_input(
"Sentence:", placeholder="Enter a sentence.", on_change=None
)
# TODO: Check if this is needed!
clicked = st.button("Submit")
if sent:
probs, labels = compute([sent], version=version)
prob = probs[0]
label = labels[0]
# Check if the file exists
if not os.path.exists('logs.txt'):
with open('logs.txt', 'w') as file:
pass
print(f"{sent}, {label}: {prob}")
with open("logs.txt", "a") as f:
f.write(f"{sent}, {label}: {prob}\n")
# plot
if version == "All-Models":
plot_multiples(["xlmrlarge", "serengeti", "afriberta", "afroxlmrbase", "afrolm", "BERT", "OpenLID", "GlotLID v3"], labels, probs)
else:
plot(label, prob)
with tab2:
version = st.radio(
"Choose model",
["xlmrlarge", "serengeti", "afriberta", "afroxlmrbase", "afrolm", "BERT","openlid-201", "GlotLID v3", "All-Models"],
captions=["za-XLMR-Large", "za-Serengeti", "za-AfriBERTa", "za-Afro-XLMR-BASE", "za-AfroLM", "za-BERT", "OpenLID", "GlotLID v3", "All-Models"],
index = 4,
key = 'version_tab2',
horizontal = True
)
file = st.file_uploader("Upload a file", type=["txt"])
if file is not None:
df = pd.read_csv(file, sep="¦\t¦", header=None, engine='python')
df.columns = ["Sentence"]
df.reset_index(drop=True, inplace=True)
# TODO: Run the model
df['Prob'], df["Label"] = compute(df["Sentence"].tolist(), version= version)
df['Language'] = df["Label"].apply(get_name)
# A horizontal rule
st.markdown("""---""")
chart = (
alt.Chart(df.reset_index())
.mark_area(color="darkorange", opacity=0.5)
.encode(
x=X(field="index", title="Sentence Index"),
y=Y("Prob", scale=Scale(domain=[0, 1])),
)
)
st.altair_chart(chart.interactive(), use_container_width=True)
col1, col2 = st.columns([4, 1])
with col1:
# Display the output
st.table(
df,
)
with col2:
# Add a download button
csv = convert_df(df)
st.download_button(
label=":file_folder: Download predictions as CSV",
data=csv,
file_name="GlotLID.csv",
mime="text/csv",
)
# citation()