Spaces:
Runtime error
Runtime error
upload app.py
Browse files- app.py +89 -0
- requirements.txt +6 -0
app.py
ADDED
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import BlenderbotTokenizer, BlenderbotForConditionalGeneration, pipeline
|
3 |
+
import torch
|
4 |
+
from gtts import gTTS
|
5 |
+
|
6 |
+
|
7 |
+
# Cargamos el modelo para el chat
|
8 |
+
model_name = 'facebook/blenderbot-400M-distill'
|
9 |
+
tokenizer = BlenderbotTokenizer.from_pretrained(model_name)
|
10 |
+
model = BlenderbotForConditionalGeneration.from_pretrained(model_name)
|
11 |
+
|
12 |
+
|
13 |
+
# Cargamos el traductor de ingles a español
|
14 |
+
english_model_name = "Helsinki-NLP/opus-mt-en-es"
|
15 |
+
translator_en_es = pipeline("translation", model=english_model_name)
|
16 |
+
|
17 |
+
# Cargamos el traductor de español a ingles
|
18 |
+
spanish_model_name = "Helsinki-NLP/opus-mt-es-en"
|
19 |
+
translator_es_en = pipeline("translation", model=spanish_model_name)
|
20 |
+
|
21 |
+
|
22 |
+
def take_last_tokens(inputs, note_history, history):
|
23 |
+
"""Filtrar los últimos 128 tokens"""
|
24 |
+
if inputs['input_ids'].shape[1] > 128:
|
25 |
+
inputs['input_ids'] = torch.tensor([inputs['input_ids'][0][-128:].tolist()])
|
26 |
+
inputs['attention_mask'] = torch.tensor([inputs['attention_mask'][0][-128:].tolist()])
|
27 |
+
note_history = ['</s> <s>'.join(note_history[0].split('</s> <s>')[2:])]
|
28 |
+
history = history[1:]
|
29 |
+
return inputs, note_history, history
|
30 |
+
|
31 |
+
def add_note_to_history(note, note_history):
|
32 |
+
"""Añadir una nota a la información histórica del chat"""
|
33 |
+
note_history.append(note)
|
34 |
+
note_history = '</s> <s>'.join(note_history)
|
35 |
+
return [note_history]
|
36 |
+
|
37 |
+
|
38 |
+
def predict(text, history):
|
39 |
+
history = history or []
|
40 |
+
if history:
|
41 |
+
history_useful = ['</s> <s>'.join([str(a[0])+'</s> <s>'+str(a[1]) for a in history])]
|
42 |
+
else:
|
43 |
+
history_useful = []
|
44 |
+
# Traducimos el texto ingresado a ingles
|
45 |
+
text_input = translator_es_en(text)[0]['translation_text']
|
46 |
+
|
47 |
+
# comparamos con el historial y codificamos la nueva entrada del usuario
|
48 |
+
history_useful = add_note_to_history(text_input, history_useful)
|
49 |
+
inputs = tokenizer(history_useful, return_tensors="pt")
|
50 |
+
inputs, history_useful, history = take_last_tokens(inputs, history_useful, history)
|
51 |
+
|
52 |
+
# Generar una respuesta
|
53 |
+
reply_ids = model.generate(**inputs)
|
54 |
+
response = tokenizer.batch_decode(reply_ids, skip_special_tokens=True)[0]
|
55 |
+
# sumamos la respuesta al historial del chat
|
56 |
+
history_useful = add_note_to_history(response, history_useful)
|
57 |
+
list_history = history_useful[0].split('</s> <s>')
|
58 |
+
history.append((list_history[-2], list_history[-1]))
|
59 |
+
|
60 |
+
# pasamos el resultado a la funcion get_speach para obtener el audio
|
61 |
+
spanish_text = translator_en_es(response)
|
62 |
+
result_es = spanish_text[0]['translation_text']
|
63 |
+
sound_file = 'output.wav'
|
64 |
+
tts = gTTS(result_es, lang="es", tld='com.mx')
|
65 |
+
tts.save(sound_file)
|
66 |
+
return sound_file, history
|
67 |
+
|
68 |
+
description = """
|
69 |
+
<h2 style="text-align:center">Inicia el chat con la IA que ha sido entrenada para hablar contigo sobre lo que quieras.</h2>
|
70 |
+
<h2 style="text-align:center">¡Hablemos!</h2>
|
71 |
+
"""
|
72 |
+
article = """Instrucciones:
|
73 |
+
\n1. Inserte el texto en la casilla de texto
|
74 |
+
\n2. Presionar 'Enviar' y esperar la respuesta
|
75 |
+
\n4. Para enviar otro texto borrar el actual y volver al punto 1.
|
76 |
+
|
77 |
+
El modelo usa:
|
78 |
+
- Modelo conversacional [facebook/blenderbot-400M-distill](https://huggingface.co/facebook/blenderbot-400M-distill?text=Hey+my+name+is+Julien%21+How+are+you%3F),
|
79 |
+
- Para las traducciones [Helsinki-NLP](https://huggingface.co/Helsinki-NLP)
|
80 |
+
- Para la respuesta de voz [gTTS](https://pypi.org/project/gTTS/)
|
81 |
+
/n... y mucha magia ☺
|
82 |
+
"""
|
83 |
+
|
84 |
+
gr.Interface(fn=predict,
|
85 |
+
title="ChatBot Text-to-Speach en Español",
|
86 |
+
inputs= [gr.Textbox("", max_lines = 5, label = "Inserte su texto aqui") , 'state'],
|
87 |
+
outputs = [gr.Audio(type='file', label="Respuesta de IA en forma de audio"), 'state'],
|
88 |
+
description = description ,
|
89 |
+
article = article).launch(debug=True)
|
requirements.txt
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
transformers==4.21.0
|
2 |
+
gradio==3.11.0
|
3 |
+
torch==1.12.1
|
4 |
+
sacremoses==0.0.53
|
5 |
+
sentencepiece==0.1.97
|
6 |
+
gtts
|