Spaces:
Runtime error
Runtime error
| import gradio as gr | |
| from transformers import BlenderbotTokenizer, BlenderbotForConditionalGeneration, pipeline | |
| import torch | |
| from gtts import gTTS | |
| # Cargamos el modelo para el chat | |
| model_name = 'facebook/blenderbot-400M-distill' | |
| tokenizer = BlenderbotTokenizer.from_pretrained(model_name) | |
| model = BlenderbotForConditionalGeneration.from_pretrained(model_name) | |
| # Cargamos el traductor de ingles a español | |
| english_model_name = "Helsinki-NLP/opus-mt-en-es" | |
| translator_en_es = pipeline("translation", model=english_model_name) | |
| # Cargamos el traductor de español a ingles | |
| spanish_model_name = "Helsinki-NLP/opus-mt-es-en" | |
| translator_es_en = pipeline("translation", model=spanish_model_name) | |
| def take_last_tokens(inputs, note_history, history): | |
| """Filtrar los últimos 128 tokens""" | |
| if inputs['input_ids'].shape[1] > 128: | |
| inputs['input_ids'] = torch.tensor([inputs['input_ids'][0][-128:].tolist()]) | |
| inputs['attention_mask'] = torch.tensor([inputs['attention_mask'][0][-128:].tolist()]) | |
| note_history = ['</s> <s>'.join(note_history[0].split('</s> <s>')[2:])] | |
| history = history[1:] | |
| return inputs, note_history, history | |
| def add_note_to_history(note, note_history): | |
| """Añadir una nota a la información histórica del chat""" | |
| note_history.append(note) | |
| note_history = '</s> <s>'.join(note_history) | |
| return [note_history] | |
| def predict(text, history): | |
| history = history or [] | |
| if history: | |
| history_useful = ['</s> <s>'.join([str(a[0])+'</s> <s>'+str(a[1]) for a in history])] | |
| else: | |
| history_useful = [] | |
| # Traducimos el texto ingresado a ingles | |
| text_input = translator_es_en(text)[0]['translation_text'] | |
| # comparamos con el historial y codificamos la nueva entrada del usuario | |
| history_useful = add_note_to_history(text_input, history_useful) | |
| inputs = tokenizer(history_useful, return_tensors="pt") | |
| inputs, history_useful, history = take_last_tokens(inputs, history_useful, history) | |
| # Generar una respuesta | |
| reply_ids = model.generate(**inputs) | |
| response = tokenizer.batch_decode(reply_ids, skip_special_tokens=True)[0] | |
| # sumamos la respuesta al historial del chat | |
| history_useful = add_note_to_history(response, history_useful) | |
| list_history = history_useful[0].split('</s> <s>') | |
| history.append((list_history[-2], list_history[-1])) | |
| # pasamos el resultado a gtts para obtener el audio | |
| spanish_text = translator_en_es(response) | |
| result_es = spanish_text[0]['translation_text'] | |
| sound_file = 'output.wav' | |
| tts = gTTS(result_es, lang="es", tld='com.mx') | |
| tts.save(sound_file) | |
| return sound_file, history | |
| description = """ | |
| <h2 style="text-align:center">Inicia el chat con la IA que ha sido entrenada para hablar contigo sobre lo que quieras.</h2> | |
| <h2 style="text-align:center">¡Hablemos!</h2> | |
| """ | |
| article = """Instrucciones: | |
| \n1. Inserte el texto en la casilla de texto | |
| \n2. Presionar 'Enviar' y esperar la respuesta | |
| \n4. Para enviar otro texto borrar el actual y volver al punto 1. | |
| El modelo usa: | |
| - Modelo conversacional [facebook/blenderbot-400M-distill](https://huggingface.co/facebook/blenderbot-400M-distill?text=Hey+my+name+is+Julien%21+How+are+you%3F), | |
| - Para las traducciones [Helsinki-NLP](https://huggingface.co/Helsinki-NLP) | |
| - Para la respuesta de voz [gTTS](https://pypi.org/project/gTTS/) | |
| \n... y mucha magia ☺ | |
| """ | |
| gr.Interface(fn=predict, | |
| title="ChatBot Text-to-Speach en Español", | |
| inputs= [gr.Textbox("", max_lines = 5, label = "Inserte su texto aqui") , 'state'], | |
| outputs = [gr.Audio(type='file', label="Respuesta de IA en forma de audio"), 'state'], | |
| description = description , | |
| article = article).launch(debug=True) |