Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,275 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import tensorflow as tf
|
3 |
+
from transformers import AutoTokenizer, TFAutoModelForQuestionAnswering, AutoModelForCausalLM
|
4 |
+
import gradio as gr
|
5 |
+
import re
|
6 |
+
|
7 |
+
# Check if GPU is available and use it if possible
|
8 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
9 |
+
|
10 |
+
# Load Spanish models and tokenizers
|
11 |
+
confli_model_spanish = 'salsarra/ConfliBERT-Spanish-Beto-Cased-NewsQA'
|
12 |
+
confli_model_spanish_qa = TFAutoModelForQuestionAnswering.from_pretrained(confli_model_spanish)
|
13 |
+
confli_tokenizer_spanish = AutoTokenizer.from_pretrained(confli_model_spanish)
|
14 |
+
|
15 |
+
beto_model_spanish = 'salsarra/Beto-Spanish-Cased-NewsQA'
|
16 |
+
beto_model_spanish_qa = TFAutoModelForQuestionAnswering.from_pretrained(beto_model_spanish)
|
17 |
+
beto_tokenizer_spanish = AutoTokenizer.from_pretrained(beto_model_spanish)
|
18 |
+
|
19 |
+
confli_sqac_model_spanish = 'salsarra/ConfliBERT-Spanish-Beto-Cased-SQAC'
|
20 |
+
confli_sqac_model_spanish_qa = TFAutoModelForQuestionAnswering.from_pretrained(confli_sqac_model_spanish)
|
21 |
+
confli_sqac_tokenizer_spanish = AutoTokenizer.from_pretrained(confli_sqac_model_spanish)
|
22 |
+
|
23 |
+
beto_sqac_model_spanish = 'salsarra/Beto-Spanish-Cased-SQAC'
|
24 |
+
beto_sqac_model_spanish_qa = TFAutoModelForQuestionAnswering.from_pretrained(beto_sqac_model_spanish)
|
25 |
+
beto_sqac_tokenizer_spanish = AutoTokenizer.from_pretrained(beto_sqac_model_spanish)
|
26 |
+
|
27 |
+
# Load Spanish GPT-2 model and tokenizer
|
28 |
+
gpt2_spanish_model_name = 'datificate/gpt2-small-spanish'
|
29 |
+
gpt2_spanish_tokenizer = AutoTokenizer.from_pretrained(gpt2_spanish_model_name)
|
30 |
+
gpt2_spanish_model = AutoModelForCausalLM.from_pretrained(gpt2_spanish_model_name).to(device)
|
31 |
+
|
32 |
+
# Load BLOOM-1.7B model and tokenizer for Spanish
|
33 |
+
bloom_model_name = 'bigscience/bloom-1b7'
|
34 |
+
bloom_tokenizer = AutoTokenizer.from_pretrained(bloom_model_name)
|
35 |
+
bloom_model = AutoModelForCausalLM.from_pretrained(bloom_model_name).to(device)
|
36 |
+
|
37 |
+
# Preload models with a dummy pass to improve first-time loading
|
38 |
+
def preload_models():
|
39 |
+
dummy_context = "Este es un contexto de prueba."
|
40 |
+
dummy_question = "驴Cu谩l es el prop贸sito de este contexto?"
|
41 |
+
|
42 |
+
# Run each model with a dummy input to initialize them
|
43 |
+
inputs = confli_tokenizer_spanish(dummy_question, dummy_context, return_tensors='tf')
|
44 |
+
_ = confli_model_spanish_qa(inputs)
|
45 |
+
|
46 |
+
inputs = beto_tokenizer_spanish(dummy_question, dummy_context, return_tensors='tf')
|
47 |
+
_ = beto_model_spanish_qa(inputs)
|
48 |
+
|
49 |
+
inputs = confli_sqac_tokenizer_spanish(dummy_question, dummy_context, return_tensors='tf')
|
50 |
+
_ = confli_sqac_model_spanish_qa(inputs)
|
51 |
+
|
52 |
+
inputs = beto_sqac_tokenizer_spanish(dummy_question, dummy_context, return_tensors='tf')
|
53 |
+
_ = beto_sqac_model_spanish_qa(inputs)
|
54 |
+
|
55 |
+
preload_models() # Initialize models
|
56 |
+
|
57 |
+
# Error handling function
|
58 |
+
def handle_error_message(e, default_limit=512):
|
59 |
+
error_message = str(e)
|
60 |
+
pattern = re.compile(r"The size of tensor a \((\d+)\) must match the size of tensor b \((\d+)\)")
|
61 |
+
match = pattern.search(error_message)
|
62 |
+
if match:
|
63 |
+
number_1, number_2 = match.groups()
|
64 |
+
return f"<span style='color: red; font-weight: bold;'>Error: Text Input is over limit where inserted text size {number_1} is larger than model limits of {number_2}</span>"
|
65 |
+
return f"<span style='color: red; font-weight: bold;'>Error: Text Input is over limit where inserted text size is larger than model limits of {default_limit}</span>"
|
66 |
+
|
67 |
+
# Spanish QA functions
|
68 |
+
def question_answering_spanish(context, question):
|
69 |
+
try:
|
70 |
+
inputs = confli_tokenizer_spanish(question, context, return_tensors='tf', truncation=True)
|
71 |
+
outputs = confli_model_spanish_qa(inputs)
|
72 |
+
answer_start = tf.argmax(outputs.start_logits, axis=1).numpy()[0]
|
73 |
+
answer_end = tf.argmax(outputs.end_logits, axis=1).numpy()[0] + 1
|
74 |
+
answer = confli_tokenizer_spanish.convert_tokens_to_string(confli_tokenizer_spanish.convert_ids_to_tokens(inputs['input_ids'].numpy()[0][answer_start:answer_end]))
|
75 |
+
return f"<span style='color: green; font-weight: bold;'>{answer}</span>"
|
76 |
+
except Exception as e:
|
77 |
+
return handle_error_message(e)
|
78 |
+
|
79 |
+
def beto_question_answering_spanish(context, question):
|
80 |
+
try:
|
81 |
+
inputs = beto_tokenizer_spanish(question, context, return_tensors='tf', truncation=True)
|
82 |
+
outputs = beto_model_spanish_qa(inputs)
|
83 |
+
answer_start = tf.argmax(outputs.start_logits, axis=1).numpy()[0]
|
84 |
+
answer_end = tf.argmax(outputs.end_logits, axis=1).numpy()[0] + 1
|
85 |
+
answer = beto_tokenizer_spanish.convert_tokens_to_string(beto_tokenizer_spanish.convert_ids_to_tokens(inputs['input_ids'].numpy()[0][answer_start:answer_end]))
|
86 |
+
return f"<span style='color: blue; font-weight: bold;'>{answer}</span>"
|
87 |
+
except Exception as e:
|
88 |
+
return handle_error_message(e)
|
89 |
+
|
90 |
+
def confli_sqac_question_answering_spanish(context, question):
|
91 |
+
try:
|
92 |
+
inputs = confli_sqac_tokenizer_spanish(question, context, return_tensors='tf', truncation=True)
|
93 |
+
outputs = confli_sqac_model_spanish_qa(inputs)
|
94 |
+
answer_start = tf.argmax(outputs.start_logits, axis=1).numpy()[0]
|
95 |
+
answer_end = tf.argmax(outputs.end_logits, axis=1).numpy()[0] + 1
|
96 |
+
answer = confli_sqac_tokenizer_spanish.convert_tokens_to_string(confli_sqac_tokenizer_spanish.convert_ids_to_tokens(inputs['input_ids'].numpy()[0][answer_start:answer_end]))
|
97 |
+
return f"<span style='color: teal; font-weight: bold;'>{answer}</span>"
|
98 |
+
except Exception as e:
|
99 |
+
return handle_error_message(e)
|
100 |
+
|
101 |
+
def beto_sqac_question_answering_spanish(context, question):
|
102 |
+
try:
|
103 |
+
inputs = beto_sqac_tokenizer_spanish(question, context, return_tensors='tf', truncation=True)
|
104 |
+
outputs = beto_sqac_model_spanish_qa(inputs)
|
105 |
+
answer_start = tf.argmax(outputs.start_logits, axis=1).numpy()[0]
|
106 |
+
answer_end = tf.argmax(outputs.end_logits, axis=1).numpy()[0] + 1
|
107 |
+
answer = beto_sqac_tokenizer_spanish.convert_tokens_to_string(beto_sqac_tokenizer_spanish.convert_ids_to_tokens(inputs['input_ids'].numpy()[0][answer_start:answer_end]))
|
108 |
+
return f"<span style='color: brown; font-weight: bold;'>{answer}</span>"
|
109 |
+
except Exception as e:
|
110 |
+
return handle_error_message(e)
|
111 |
+
|
112 |
+
def gpt2_spanish_question_answering(context, question):
|
113 |
+
try:
|
114 |
+
prompt = f"Contexto:\n{context}\n\nPregunta:\n{question}\n\nRespuesta:"
|
115 |
+
inputs = gpt2_spanish_tokenizer(prompt, return_tensors='pt').to(device)
|
116 |
+
outputs = gpt2_spanish_model.generate(
|
117 |
+
inputs['input_ids'],
|
118 |
+
max_length=inputs['input_ids'].shape[1] + 50,
|
119 |
+
num_return_sequences=1,
|
120 |
+
pad_token_id=gpt2_spanish_tokenizer.eos_token_id,
|
121 |
+
do_sample=True,
|
122 |
+
top_k=40,
|
123 |
+
temperature=0.8
|
124 |
+
)
|
125 |
+
answer = gpt2_spanish_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
126 |
+
answer = answer.split("Respuesta:")[-1].strip()
|
127 |
+
return f"<span style='color: orange; font-weight: bold;'>{answer}</span>"
|
128 |
+
except Exception as e:
|
129 |
+
return handle_error_message(e)
|
130 |
+
|
131 |
+
def bloom_question_answering(context, question):
|
132 |
+
try:
|
133 |
+
prompt = f"Contexto:\n{context}\n\nPregunta:\n{question}\n\nRespuesta:"
|
134 |
+
inputs = bloom_tokenizer(prompt, return_tensors='pt').to(device)
|
135 |
+
outputs = bloom_model.generate(
|
136 |
+
inputs['input_ids'],
|
137 |
+
max_length=inputs['input_ids'].shape[1] + 50,
|
138 |
+
num_return_sequences=1,
|
139 |
+
pad_token_id=bloom_tokenizer.eos_token_id,
|
140 |
+
do_sample=True,
|
141 |
+
top_k=40,
|
142 |
+
temperature=0.8
|
143 |
+
)
|
144 |
+
answer = bloom_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
145 |
+
answer = answer.split("Respuesta:")[-1].strip()
|
146 |
+
return f"<span style='color: purple; font-weight: bold;'>{answer}</span>"
|
147 |
+
except Exception as e:
|
148 |
+
return handle_error_message(e)
|
149 |
+
|
150 |
+
# Main function for Spanish QA
|
151 |
+
def compare_question_answering_spanish(context, question):
|
152 |
+
confli_answer_spanish = question_answering_spanish(context, question)
|
153 |
+
beto_answer_spanish = beto_question_answering_spanish(context, question)
|
154 |
+
confli_sqac_answer_spanish = confli_sqac_question_answering_spanish(context, question)
|
155 |
+
beto_sqac_answer_spanish = beto_sqac_question_answering_spanish(context, question)
|
156 |
+
gpt2_answer_spanish = gpt2_spanish_question_answering(context, question)
|
157 |
+
bloom_answer = bloom_question_answering(context, question)
|
158 |
+
return f"""
|
159 |
+
<div>
|
160 |
+
<h2 style='color: #2e8b57; font-weight: bold;'>Respuestas:</h2>
|
161 |
+
</div><br>
|
162 |
+
<div>
|
163 |
+
<strong>ConfliBERT-Spanish-Beto-Cased-NewsQA:</strong><br>{confli_answer_spanish}</div><br>
|
164 |
+
<div>
|
165 |
+
<strong>Beto-Spanish-Cased-NewsQA:</strong><br>{beto_answer_spanish}
|
166 |
+
</div><br>
|
167 |
+
<div>
|
168 |
+
<strong>ConfliBERT-Spanish-Beto-Cased-SQAC:</strong><br>{confli_sqac_answer_spanish}
|
169 |
+
</div><br>
|
170 |
+
<div>
|
171 |
+
<strong>Beto-Spanish-Cased-SQAC:</strong><br>{beto_sqac_answer_spanish}
|
172 |
+
</div><br>
|
173 |
+
<div>
|
174 |
+
<strong>GPT-2-Small-Spanish:</strong><br>{gpt2_answer_spanish}
|
175 |
+
</div><br>
|
176 |
+
<div>
|
177 |
+
<strong>BLOOM-1.7B:</strong><br>{bloom_answer}
|
178 |
+
</div><br>
|
179 |
+
<div>
|
180 |
+
<strong>Informaci贸n del modelo:</strong><br>
|
181 |
+
ConfliBERT-Spanish-Beto-Cased-NewsQA: <a href='https://huggingface.co/salsarra/ConfliBERT-Spanish-Beto-Cased-NewsQA' target='_blank'>salsarra/ConfliBERT-Spanish-Beto-Cased-NewsQA</a><br>
|
182 |
+
Beto-Spanish-Cased-NewsQA: <a href='https://huggingface.co/salsarra/Beto-Spanish-Cased-NewsQA' target='_blank'>salsarra/Beto-Spanish-Cased-NewsQA</a><br>
|
183 |
+
ConfliBERT-Spanish-Beto-Cased-SQAC: <a href='https://huggingface.co/salsarra/ConfliBERT-Spanish-Beto-Cased-SQAC' target='_blank'>salsarra/ConfliBERT-Spanish-Beto-Cased-SQAC</a><br>
|
184 |
+
Beto-Spanish-Cased-SQAC: <a href='https://huggingface.co/salsarra/Beto-Spanish-Cased-SQAC' target='_blank'>salsarra/Beto-Spanish-Cased-SQAC</a><br>
|
185 |
+
GPT-2-Small-Spanish: <a href='https://huggingface.co/datificate/gpt2-small-spanish' target='_blank'>datificate GPT-2 Small Spanish</a><br>
|
186 |
+
BLOOM-1.7B: <a href='https://huggingface.co/bigscience/bloom-1b7' target='_blank'>bigscience BLOOM-1.7B</a><br>
|
187 |
+
</div>
|
188 |
+
"""
|
189 |
+
|
190 |
+
# CSS for Gradio interface
|
191 |
+
css_styles = """
|
192 |
+
body {
|
193 |
+
background-color: #f0f8ff;
|
194 |
+
font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif;
|
195 |
+
}
|
196 |
+
h1 a {
|
197 |
+
color: #2e8b57;
|
198 |
+
text-align: center;
|
199 |
+
font-size: 2em;
|
200 |
+
text-decoration: none;
|
201 |
+
}
|
202 |
+
h1 a:hover {
|
203 |
+
color: #ff8c00;
|
204 |
+
}
|
205 |
+
h2 {
|
206 |
+
color: #ff8c00;
|
207 |
+
text-align: center;
|
208 |
+
font-size: 1.5em;
|
209 |
+
}
|
210 |
+
.description-light {
|
211 |
+
color: black;
|
212 |
+
display: block;
|
213 |
+
font-size: 1em;
|
214 |
+
text-align: center;
|
215 |
+
}
|
216 |
+
.description-dark {
|
217 |
+
color: white;
|
218 |
+
display: none;
|
219 |
+
font-size: 1em;
|
220 |
+
text-align: center;
|
221 |
+
}
|
222 |
+
@media (prefers-color-scheme: dark) {
|
223 |
+
.description-light {
|
224 |
+
display: none;
|
225 |
+
}
|
226 |
+
.description-dark {
|
227 |
+
display: block;
|
228 |
+
}
|
229 |
+
}
|
230 |
+
.footer {
|
231 |
+
text-align: center;
|
232 |
+
margin-top: 10px;
|
233 |
+
font-size: 0.9em;
|
234 |
+
color: #666;
|
235 |
+
width: 100%;
|
236 |
+
}
|
237 |
+
.footer a {
|
238 |
+
color: #2e8b57;
|
239 |
+
font-weight: bold;
|
240 |
+
text-decoration: none;
|
241 |
+
}
|
242 |
+
.footer a:hover {
|
243 |
+
text-decoration: underline;
|
244 |
+
}
|
245 |
+
"""
|
246 |
+
|
247 |
+
# Define the Gradio interface with footer directly in the layout
|
248 |
+
demo = gr.Interface(
|
249 |
+
fn=compare_question_answering_spanish,
|
250 |
+
inputs=[
|
251 |
+
gr.Textbox(lines=5, placeholder="Ingrese el contexto aqu铆...", label="Contexto"),
|
252 |
+
gr.Textbox(lines=2, placeholder="Ingrese su pregunta aqu铆...", label="Pregunta")
|
253 |
+
],
|
254 |
+
outputs=gr.HTML(label="Salida"),
|
255 |
+
title="<a href='https://eventdata.utdallas.edu/conflibert/' target='_blank'>ConfliBERT-Spanish-QA</a>",
|
256 |
+
description="""
|
257 |
+
<span class="description-light">Compare respuestas entre los modelos ConfliBERT, BETO, ConfliBERT SQAC, Beto SQAC, GPT-2 Small Spanish y BLOOM-1.7B para preguntas en espa帽ol.</span>
|
258 |
+
<span class="description-dark">Compare respuestas entre los modelos ConfliBERT, BETO, ConfliBERT SQAC, Beto SQAC, GPT-2 Small Spanish y BLOOM-1.7B para preguntas en espa帽ol.</span>
|
259 |
+
""",
|
260 |
+
css=css_styles,
|
261 |
+
allow_flagging="never",
|
262 |
+
# Footer HTML with centered, green links
|
263 |
+
article="""
|
264 |
+
<div class='footer'>
|
265 |
+
<a href='https://eventdata.utdallas.edu/' style='color: #2e8b57; font-weight: bold;'>UTD Event Data</a> |
|
266 |
+
<a href='https://www.utdallas.edu/' style='color: #2e8b57; font-weight: bold;'>University of Texas at Dallas</a>
|
267 |
+
</div>
|
268 |
+
<div class='footer'>
|
269 |
+
Developed By: <a href='https://www.linkedin.com/in/sultan-alsarra-phd-56977a63/' target='_blank' style='color: #2e8b57; font-weight: bold;'>Sultan Alsarra</a>
|
270 |
+
</div>
|
271 |
+
"""
|
272 |
+
)
|
273 |
+
|
274 |
+
# Launch the Gradio demo
|
275 |
+
demo.launch(share=True)
|