How It Works¶

On startup the demo application reads command line parameters and loads a network to Inference engine. It also fetches data from the user-provided url to populate the “context” text. The text is then used to search answers for user-provided questions.

Preparing to Run¶

The list of models supported by the demo is in <omz_dir>/demos/bert_question_answering_demo/python/models.lst file. This file can be used as a parameter for Model Downloader and Converter to download and, if necessary, convert models to OpenVINO Inference Engine format (*.xml + *.bin).

python3 <omz_dir>/tools/downloader/downloader.py --list models.lst

An example of using the Model Converter:

python3 <omz_dir>/tools/downloader/converter.py --list models.lst

Supported Models¶

The “small” variants of these are so-called “distilled” models, which originated from the BERT Large but substantially smaller and faster. If you want to use an official MLPerf* BERT ONNX model rather than the distilled model on the Open model Zoo, the command line to convert the int8 model is as follows:

python3 mo.py
-m <path_to_model>/bert_large_v1_1_fake_quant.onnx
--input_shape "[1,384],[1,384],[1,384]"
--keep_shape_ops

Note

Refer to the tables Intel’s Pre-Trained Models Device Support and Public Pre-Trained Models Device Support for the details on models inference support at different devices.

Running¶

Running the application with the -h option yields the following usage message:

usage: bert_question_answering_demo.py [-h] -v VOCAB -m MODEL -i INPUT
[--questions QUESTION [QUESTION ...]]
[--input_names INPUT_NAMES]
[--output_names OUTPUT_NAMES]
[-q MAX_QUESTION_TOKEN_NUM]
[-r] [-c]

Options:
-h, --help            Show this help message and exit.
-v VOCAB, --vocab VOCAB
Required. path to the vocabulary file with tokens
-m MODEL, --model MODEL
Required. Path to an .xml file with a trained model
-i INPUT, --input INPUT
Required. URL to a page with context
--questions QUESTION [QUESTION ...]
Optional. Prepared questions
--input_names INPUT_NAMES
Optional. Inputs names for the network. Default values
--output_names OUTPUT_NAMES
Optional. Outputs names for the network. Default
values are "output_s,output_e"
Optional. SQUAD version used for model fine tuning
-q MAX_QUESTION_TOKEN_NUM, --max_question_token_num MAX_QUESTION_TOKEN_NUM
Optional. Maximum number of tokens in question
Optional. Maximum number of tokens in answer
-d DEVICE, --device DEVICE
Optional. Target device to perform inference
on.Default value is CPU
-r, --reshape         Optional. Auto reshape sequence length to the input
context + max question len (to improve the speed)
-c, --colors          Optional. Nice coloring of the questions/answers.
Might not work on some terminals (like Windows* cmd
console)

Example Demo Cmd-Line¶

python3 bert_question_answering_demo.py
--output_names="output_s,output_e"
--input="https://en.wikipedia.org/wiki/Bert_(Sesame_Street)"
-c

The demo will use a wiki-page about the Bert character to answer your questions like “who is Bert”, “how old is Bert”, etc.

Demo Inputs¶

The application reads text from the HTML page at the given url and then answers questions typed from the console. The model and its parameters (inputs and outputs) are also important demo arguments. Notice that since order of inputs for the model does matter, the demo script checks that the inputs specified from the command-line match the actual network inputs. When the reshape option (-r) is specified, the script also attempts to reshape the network to the length of the context plus length of the question (both in tokens), if the resulting value is smaller than the original sequence length that the network expects. This is performance (speed) and memory footprint saving option. Since some networks are not-reshapable (due to limitations of the internal layers) the reshaping might fail, so you will need to run the demo without it. Please see general reshape intro and limitations

Demo Outputs¶

The application outputs found answers to the same console.

Classifying Documents with Long Texts¶

Notice that when the original “context” (text from the url) together with the question do not fit the model input (usually 384 tokens for the Bert-Large, or 128 for the Bert-Base), the demo splits the context into overlapping segments. Thus, for the long texts, the network is called multiple times. The results are then sorted by the probabilities.

Demo Performance¶

Even though the demo reports inference performance (by measuring wall-clock time for individual inference calls), it is only baseline performance, as certain tricks like batching, throughput mode can be applied. Please use the full-blown Benchmark C++ Sample for any actual performance measurements.