Spaces:
Runtime error
Runtime error
import gradio as gr | |
from transformers import ( | |
AutoTokenizer, | |
AutoModelForSeq2SeqLM, | |
AutoProcessor, | |
AutoModelForDocumentQuestionAnswering, | |
pipeline, | |
) | |
import torch | |
import numpy as np | |
processor = AutoProcessor.from_pretrained( | |
"MariaK/layoutlmv2-base-uncased_finetuned_docvqa_v2" | |
) | |
model = AutoModelForDocumentQuestionAnswering.from_pretrained( | |
"MariaK/layoutlmv2-base-uncased_finetuned_docvqa_v2" | |
) | |
tokenizer_ru2en = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ru-en") | |
model_ru2en = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-ru-en") | |
tokenizer_en2ru = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-ru") | |
model_en2ru = AutoModelForSeq2SeqLM.from_pretrained("Helsinki-NLP/opus-mt-en-ru") | |
transcriber = pipeline( | |
"automatic-speech-recognition", model="artyomboyko/whisper-base-fine_tuned-ru" | |
) | |
def translate_ru2en(text): | |
inputs = tokenizer_ru2en(text, return_tensors="pt") | |
outputs = model_ru2en.generate(**inputs) | |
translated_text = tokenizer_ru2en.decode(outputs[0], skip_special_tokens=True) | |
return translated_text | |
def translate_en2ru(text): | |
inputs = tokenizer_en2ru(text, return_tensors="pt") | |
outputs = model_en2ru.generate(**inputs) | |
translated_text = tokenizer_en2ru.decode(outputs[0], skip_special_tokens=True) | |
return translated_text | |
def generate_answer_git(image, question): | |
with torch.no_grad(): | |
encoding = processor( | |
images=image, | |
text=question, | |
return_tensors="pt", | |
max_length=512, | |
truncation=True, | |
) | |
outputs = model(**encoding) | |
start_logits = outputs.start_logits | |
end_logits = outputs.end_logits | |
predicted_start_idx = start_logits.argmax(-1).item() | |
predicted_end_idx = end_logits.argmax(-1).item() | |
return processor.tokenizer.decode( | |
encoding.input_ids.squeeze()[predicted_start_idx : predicted_end_idx + 1] | |
) | |
def generate_answer(image, question): | |
question_en = translate_ru2en(question) | |
print(f"Вопрос на английском: {question_en}") | |
answer_en = generate_answer_git(image, question_en) | |
print(f"Ответ на английском: {answer_en}") | |
answer_ru = translate_en2ru(answer_en) | |
return answer_ru | |
def transcribe(image, audio): | |
if not image or not audio: | |
return | |
sr, y = audio | |
# Convert to mono if stereo | |
if y.ndim > 1: | |
y = y.mean(axis=1) | |
y = y.astype(np.float32) | |
y /= np.max(np.abs(y)) | |
return generate_answer(image, transcriber({"sampling_rate": sr, "raw": y})["text"]) | |
qa_interface = gr.Interface( | |
fn=generate_answer, | |
inputs=[ | |
gr.Image(type="pil"), | |
gr.Textbox(label="Вопрос (на русском)", placeholder="Ваш вопрос"), | |
], | |
outputs=gr.Textbox(label="Ответ (на русском)"), | |
examples=[["doc.png", "О чем данный документ?"]], | |
live=False, | |
) | |
# Interface for real-time speech recognition | |
speech_interface = gr.Interface( | |
fn=transcribe, | |
inputs=[ | |
gr.Image(type="pil"), | |
gr.Audio(sources="microphone", label="Голосовой ввод"), | |
], | |
outputs=gr.Textbox(label="Распознанный текст"), | |
live=True, | |
) | |
# Combine the interfaces in a Gradio Tabbed layout | |
interface = gr.TabbedInterface( | |
[qa_interface, speech_interface], | |
["Текстовый вопрос", "Голосовой вопрос"], | |
title="Демо визуального ответчика на вопросы (на русском)", | |
# description=( | |
# "Gradio демо для модели doc-qa с переводом вопросов и ответов" | |
# "на русский язык. Загрузите изображение и задайте вопрос, чтобы" | |
# "получить ответ. Вы также можете использовать голосовой ввод!" | |
# ), | |
# live=True, | |
) | |
interface.launch(debug=True, share=True) | |