Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import streamlit as st
|
| 2 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
| 3 |
+
import pandas as pd
|
| 4 |
+
|
| 5 |
+
# Charger le modèle et le tokenizer
|
| 6 |
+
checkpoint = "Propicto/t2p-t5-large-orfeo"
|
| 7 |
+
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
|
| 8 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(checkpoint)
|
| 9 |
+
|
| 10 |
+
# Lire le lexique
|
| 11 |
+
@st.cache
|
| 12 |
+
def read_lexicon(lexicon):
|
| 13 |
+
df = pd.read_csv(lexicon, sep='\t')
|
| 14 |
+
df['keyword_no_cat'] = df['lemma'].str.split(' #').str[0].str.strip().str.replace(' ', '_')
|
| 15 |
+
return df
|
| 16 |
+
|
| 17 |
+
lexicon = read_lexicon("lexicon.csv")
|
| 18 |
+
|
| 19 |
+
# Processus de sortie de la traduction
|
| 20 |
+
def process_output_trad(pred):
|
| 21 |
+
return pred.split()
|
| 22 |
+
|
| 23 |
+
def get_id_picto_from_predicted_lemma(df_lexicon, lemma):
|
| 24 |
+
id_picto = df_lexicon.loc[df_lexicon['keyword_no_cat'] == lemma, 'id_picto'].tolist()
|
| 25 |
+
return (id_picto[0], lemma) if id_picto else (0, lemma)
|
| 26 |
+
|
| 27 |
+
# Génération du contenu HTML pour afficher les pictogrammes
|
| 28 |
+
def generate_html(ids):
|
| 29 |
+
html_content = '<html><body>'
|
| 30 |
+
for picto_id, lemma in ids:
|
| 31 |
+
if picto_id != 0: # ignore invalid IDs
|
| 32 |
+
img_url = f"https://static.arasaac.org/pictograms/{picto_id}/{picto_id}_500.png"
|
| 33 |
+
html_content += f'''
|
| 34 |
+
<figure style="display:inline-block; margin:1px;">
|
| 35 |
+
<img src="{img_url}" alt="{lemma}" width="200" height="200" />
|
| 36 |
+
<figcaption>{lemma}</figcaption>
|
| 37 |
+
</figure>
|
| 38 |
+
'''
|
| 39 |
+
html_content += '</body></html>'
|
| 40 |
+
return html_content
|
| 41 |
+
|
| 42 |
+
# Interface utilisateur
|
| 43 |
+
st.title("Pictogramme Générateur de Traduction")
|
| 44 |
+
|
| 45 |
+
sentence = st.text_input("Entrez une phrase en français:")
|
| 46 |
+
if sentence:
|
| 47 |
+
inputs = tokenizer(sentence, return_tensors="pt").input_ids
|
| 48 |
+
outputs = model.generate(inputs, max_new_tokens=40, do_sample=True, top_k=30, top_p=0.95)
|
| 49 |
+
pred = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 50 |
+
|
| 51 |
+
sentence_to_map = process_output_trad(pred)
|
| 52 |
+
pictogram_ids = [get_id_picto_from_predicted_lemma(lexicon, lemma) for lemma in sentence_to_map]
|
| 53 |
+
|
| 54 |
+
html = generate_html(pictogram_ids)
|
| 55 |
+
st.components.v1.html(html, height=600, scrolling=True)
|