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