213-question-answering: Interactive Question Answering with OpenVINO


Fri Jun 17 2022 05:03:05 GMT+0000 (UTC)

Saved by @OpenVINOtoolkit #python #openvino #openvino-notebooks #deeplearning #accelerated-inference ##nlp #question-answering #bert

# Imports
import operator
import time
from urllib import parse

import numpy as np
from openvino.runtime import Core

import html_reader as reader
import tokens_bert as tokens

# Download the model
# directory where model will be downloaded
base_model_dir = "model"

# desired precision
precision = "FP16-INT8"

# model name as named in Open Model Zoo
model_name = "bert-small-uncased-whole-word-masking-squad-int8-0002"

model_path = f"model/intel/{model_name}/{precision}/{model_name}.xml"
model_weights_path = f"model/intel/{model_name}/{precision}/{model_name}.bin"

download_command = f"omz_downloader " \
                   f"--name {model_name} " \
                   f"--precision {precision} " \
                   f"--output_dir {base_model_dir} " \
                   f"--cache_dir {base_model_dir}"
! $download_command

# Load the model
# initialize inference engine
core = Core()
# read the network and corresponding weights from file
model = core.read_model(model=model_path, weights=model_weights_path)
# load the model on the CPU (you can use GPU as well)
compiled_model = core.compile_model(model=model, device_name="CPU")

# get input and output names of nodes
input_keys = list(compiled_model.inputs)
output_keys = list(compiled_model.outputs)

# get network input size
input_size = compiled_model.input(0).shape[1]

# Processing
# path to vocabulary file
vocab_file_path = "data/vocab.txt"

# create dictionary with words and their indices
vocab = tokens.load_vocab_file(vocab_file_path)

# define special tokens
cls_token = vocab["[CLS]"]
pad_token = vocab["[PAD]"]
sep_token = vocab["[SEP]"]

# function to load text from given urls
def load_context(sources):
    input_urls = []
    paragraphs = []
    for source in sources:
        result = parse.urlparse(source)
        if all([result.scheme, result.netloc]):

    # produce one big context string
    return "\n".join(paragraphs)

# Preprocessing
# generator of a sequence of inputs
def prepare_input(question_tokens, context_tokens):
    # length of question in tokens
    question_len = len(question_tokens)
    # context part size
    context_len = input_size - question_len - 3

    if context_len < 16:
        raise RuntimeError("Question is too long in comparison to input size. No space for context")

    # take parts of context with overlapping by 0.5
    for start in range(0, max(1, len(context_tokens) - context_len), context_len // 2):
        # part of context
        part_context_tokens = context_tokens[start:start + context_len]
        # input: question and context separated by special tokens
        input_ids = [cls_token] + question_tokens + [sep_token] + part_context_tokens + [sep_token]
        # 1 for any index if there is no padding token, 0 otherwise
        attention_mask = [1] * len(input_ids)
        # 0 for question tokens, 1 for context part
        token_type_ids = [0] * (question_len + 2) + [1] * (len(part_context_tokens) + 1)

        # add padding at the end
        (input_ids, attention_mask, token_type_ids), pad_number = pad(input_ids=input_ids,

        # create input to feed the model
        input_dict = {
            "input_ids": np.array([input_ids], dtype=np.int32),
            "attention_mask": np.array([attention_mask], dtype=np.int32),
            "token_type_ids": np.array([token_type_ids], dtype=np.int32),

        # some models require additional position_ids
        if "position_ids" in [i_key.any_name for i_key in input_keys]:
            position_ids = np.arange(len(input_ids))
            input_dict["position_ids"] = np.array([position_ids], dtype=np.int32)

        yield input_dict, pad_number, start

# function to add padding
def pad(input_ids, attention_mask, token_type_ids):
    # how many padding tokens
    diff_input_size = input_size - len(input_ids)

    if diff_input_size > 0:
        # add padding to all inputs
        input_ids = input_ids + [pad_token] * diff_input_size
        attention_mask = attention_mask + [0] * diff_input_size
        token_type_ids = token_type_ids + [0] * diff_input_size

    return (input_ids, attention_mask, token_type_ids), diff_input_size

# Postprocessing
# based on https://github.com/openvinotoolkit/open_model_zoo/blob/bf03f505a650bafe8da03d2747a8b55c5cb2ef16/demos/common/python/openvino/model_zoo/model_api/models/bert.py#L163
def postprocess(output_start, output_end, question_tokens, context_tokens_start_end, padding, start_idx):

    def get_score(logits):
        out = np.exp(logits)
        return out / out.sum(axis=-1)

    # get start-end scores for context
    score_start = get_score(output_start)
    score_end = get_score(output_end)

    # index of first context token in tensor
    context_start_idx = len(question_tokens) + 2
    # index of last+1 context token in tensor
    context_end_idx = input_size - padding - 1

    # find product of all start-end combinations to find the best one
    max_score, max_start, max_end = find_best_answer_window(start_score=score_start,

    # convert to context text start-end index
    max_start = context_tokens_start_end[max_start + start_idx][0]
    max_end = context_tokens_start_end[max_end + start_idx][1]

    return max_score, max_start, max_end

# based on https://github.com/openvinotoolkit/open_model_zoo/blob/bf03f505a650bafe8da03d2747a8b55c5cb2ef16/demos/common/python/openvino/model_zoo/model_api/models/bert.py#L188
def find_best_answer_window(start_score, end_score, context_start_idx, context_end_idx):
    context_len = context_end_idx - context_start_idx
    score_mat = np.matmul(
        start_score[context_start_idx:context_end_idx].reshape((context_len, 1)),
        end_score[context_start_idx:context_end_idx].reshape((1, context_len)),
    # reset candidates with end before start
    score_mat = np.triu(score_mat)
    # reset long candidates (>16 words)
    score_mat = np.tril(score_mat, 16)
    # find the best start-end pair
    max_s, max_e = divmod(score_mat.flatten().argmax(), score_mat.shape[1])
    max_score = score_mat[max_s, max_e]

    return max_score, max_s, max_e

def get_best_answer(question, context):
    # convert context string to tokens
    context_tokens, context_tokens_start_end = tokens.text_to_tokens(text=context.lower(),
    # convert question string to tokens
    question_tokens, _ = tokens.text_to_tokens(text=question.lower(), vocab=vocab)

    results = []
    # iterate through different parts of context
    for network_input, padding, start_idx in prepare_input(question_tokens=question_tokens,
        # get output layers
        output_start_key = compiled_model.output("output_s")
        output_end_key = compiled_model.output("output_e")

        # openvino inference
        result = compiled_model(network_input)
        # postprocess the result getting the score and context range for the answer
        score_start_end = postprocess(output_start=result[output_start_key][0],

    # find the highest score
    answer = max(results, key=operator.itemgetter(0))
    # return the part of the context, which is already an answer
    return context[answer[1]:answer[2]], answer[0]

# Main Processing Function
def run_question_answering(sources):
    print(f"Context: {sources}", flush=True)
    context = load_context(sources)

    if len(context) == 0:
        print("Error: Empty context or outside paragraphs")

    while True:
        question = input()
        # if no question - break
        if question == "":

        # measure processing time
        start_time = time.perf_counter()
        answer, score = get_best_answer(question=question, context=context)
        end_time = time.perf_counter()

        print(f"Question: {question}")
        print(f"Answer: {answer}")
        print(f"Score: {score:.2f}")
        print(f"Time: {end_time - start_time:.2f}s")

# Run on local paragraphs
sources = ["Computational complexity theory is a branch of the theory of computation in theoretical computer "
           "science that focuses on classifying computational problems according to their inherent difficulty, "
           "and relating those classes to each other. A computational problem is understood to be a task that "
           "is in principle amenable to being solved by a computer, which is equivalent to stating that the "
           "problem may be solved by mechanical application of mathematical steps, such as an algorithm."]


# Run on websites
sources = ["https://en.wikipedia.org/wiki/OpenVINO"]


Question answering is a part of information retrieval and natural language processing (NLP), which is able to answer questions posed by humans in a natural language. Question answering systems can pull answers from an unstructured collection of natural language documents called, in that case, knowledge base. This notebook demonstrates question answering with OpenVINO using the Squad-tuned BERT model from Open Model Zoo. This demo shows interactive question answering with OpenVINO. We use small BERT-large-like model distilled and quantized to INT8 on SQuAD v1.1 training set from larger BERT-large model. The model comes from Open Model Zoo. At the bottom of this notebook, you will see live inference results from your inputs. If you have not yet installed OpenVINO™, please follow the Installation Guide to install all required dependencies. https://github.com/openvinotoolkit/openvino_notebooks/blob/main/README.md#-installation-guide Link to .bin and .xml files: https://github.com/openvinotoolkit/open_model_zoo/blob/master/models/intel/bert-small-uncased-whole-word-masking-squad-int8-0002/model.yml