tests/tools/cli/test_quantize_cli.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-
#
# Copyright (c) 2022 Intel Corporation
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# SPDX-License-Identifier: Apache-2.0
#

import os
import pytest
import shutil
import tempfile

from click.testing import CliRunner
from pathlib import Path
from unittest.mock import MagicMock, patch
from tlt.tools.cli.commands.quantize import quantize
from tlt.utils.types import FrameworkType
from tlt.utils.file_utils import download_and_extract_zip_file


@pytest.mark.common
@pytest.mark.parametrize('model_name,framework,batch_size',
                         [['efficientnet_b0', FrameworkType.TENSORFLOW, 512],
                          ['inception_v3', FrameworkType.TENSORFLOW, 32],
                          ['resnet50', FrameworkType.PYTORCH, 128],
                          ['bert-base-cased', FrameworkType.PYTORCH, 256]])
@patch("tlt.models.model_factory.get_model")
@patch("tlt.datasets.dataset_factory.load_dataset")
def test_quantize(mock_load_dataset, mock_get_model, model_name, framework, batch_size):
    """
    Tests the quantize command and verifies that the
    expected calls are made on the tlt model object. The call parameters also verify that the quantize command
    is able to properly identify the model's name based on the directory and the framework type based on the
    type of saved model.
    """
    runner = CliRunner()

    tmp_dir = tempfile.mkdtemp()
    model_dir = os.path.join(tmp_dir, model_name, '3')
    dataset_dir = os.path.join(tmp_dir, 'data')
    output_dir = os.path.join(tmp_dir, 'output')

    if model_name == "bert-base-cased":
        # Get the dataset
        zip_file_url = "https://archive.ics.uci.edu/static/public/228/sms+spam+collection.zip"
        csv_dir = os.path.join(dataset_dir, "sms_spam_collection")
        csv_file_name = "SMSSpamCollection"
        delimiter = '\t'

        # If the SMS Spam collection csv file is not found, download and extract the file:
        if not os.path.exists(os.path.join(csv_dir, csv_file_name)):
            # Download the zip file with the SMS Spam collection dataset
            download_and_extract_zip_file(zip_file_url, csv_dir)

    try:
        for new_dir in [model_dir, dataset_dir]:
            os.makedirs(new_dir, exist_ok=True)

        if framework == FrameworkType.TENSORFLOW:
            Path(os.path.join(model_dir, 'saved_model.pb')).touch()
        elif framework == FrameworkType.PYTORCH:
            Path(os.path.join(model_dir, 'model.pt')).touch()

        model_mock = MagicMock()
        data_mock = MagicMock()

        if model_name == "bert-base-cased":
            model_mock.use_case = "text_classification"
        else:
            model_mock.use_case = "image_classification"

        mock_get_model.return_value = model_mock
        mock_load_dataset.return_value = data_mock

        # Call the quantize command
        if model_mock.use_case == "image_classification":
            result = runner.invoke(quantize,
                                   ["--model-dir", model_dir, "--dataset_dir", dataset_dir,
                                    "--batch-size", batch_size, "--output-dir", output_dir])
        else:
            result = runner.invoke(quantize,
                                   ["--model-dir", model_dir, "--dataset_dir", dataset_dir,
                                    "--batch-size", batch_size, "--output-dir", output_dir,
                                    "--dataset-file", csv_file_name, "--delimiter", delimiter])

        # Verify that the expected calls were made, including to create an Intel Neural Compressor config file
        mock_get_model.assert_called_once_with(model_name, framework)

        if model_mock.use_case == "image_classification":
            mock_load_dataset.assert_called_once_with(dataset_dir, model_mock.use_case, model_mock.framework)
        else:
            mock_load_dataset.assert_called_once_with(dataset_dir, model_mock.use_case, model_mock.framework,
                                                      csv_file_name=csv_file_name, delimiter=delimiter)

        assert model_mock.quantize.called

        # Verify a successful exit code
        assert result.exit_code == 0

    finally:
        if os.path.exists(tmp_dir):
            shutil.rmtree(tmp_dir)


@pytest.mark.common
@pytest.mark.parametrize('model_name,model_file',
                         [['bar', 'unsupported_model_type.txt'],
                          ['foo', 'potato.pb']])
def test_quantize_bad_model_file(model_name, model_file):
    """
    Verifies that the quantize command fails if it's given a model directory that doesn't contain a saved_model.pb or
    model.pt file.
    """
    runner = CliRunner()

    tmp_dir = tempfile.mkdtemp()
    model_dir = os.path.join(tmp_dir, model_name, '3')
    dataset_dir = os.path.join(tmp_dir, 'data')
    output_dir = os.path.join(tmp_dir, 'output')

    try:
        for new_dir in [model_dir, dataset_dir]:
            os.makedirs(new_dir)

        # Create the bogus model file
        Path(os.path.join(model_dir, model_file)).touch()

        # Call the quantize command with the bogus model directory
        result = runner.invoke(quantize,
                               ["--model-dir", model_dir, "--dataset_dir", dataset_dir, "--output-dir", output_dir])

        # Verify that we got an error about the unsupported model type
        assert result.exit_code == 1
        assert "Quantization is currently only implemented for TensorFlow saved_model.pb and PyTorch model.pt models." \
               in result.output
    finally:
        if os.path.exists(tmp_dir):
            shutil.rmtree(tmp_dir)


@pytest.mark.common
@pytest.mark.parametrize('model_name,model_file,framework',
                         [['bar', 'saved_model.pb', 'tensorflow'],
                          ['foo', 'model.pt', 'pytorch']])
def test_quantize_bad_model_dir(model_name, model_file, framework):
    """
    Verifies that quantize command fails if it's given a model directory with a model name that we don't support
    """
    runner = CliRunner()

    tmp_dir = tempfile.mkdtemp()
    model_dir = os.path.join(tmp_dir, model_name, '3')
    dataset_dir = os.path.join(tmp_dir, 'data')
    output_dir = os.path.join(tmp_dir, 'output')

    try:
        for new_dir in [model_dir, dataset_dir]:
            os.makedirs(new_dir)

        # Create the model file
        Path(os.path.join(model_dir, model_file)).touch()

        # Call the quantize command with the model directory
        result = runner.invoke(quantize,
                               ["--model-dir", model_dir, "--dataset_dir", dataset_dir, "--output-dir", output_dir])

        # Verify that we got an error about the unsupported model for the framework
        assert result.exit_code == 1
        assert "An error occurred while getting the model" in result.output
        assert "The specified model is not supported for {}".format(framework) in result.output
    finally:
        if os.path.exists(tmp_dir):
            shutil.rmtree(tmp_dir)


@pytest.mark.common
def test_quantize_model_dir_does_not_exist():
    """
    Verifies that quantize command fails if the model directory does not exist
    """
    runner = CliRunner()

    tmp_dir = tempfile.mkdtemp()
    model_dir = os.path.join(tmp_dir, 'resnet_v1_50', '3')
    dataset_dir = os.path.join(tmp_dir, 'data')
    output_dir = os.path.join(tmp_dir, 'output')

    try:
        os.makedirs(dataset_dir)

        # Call the quantize command with the model directory
        result = runner.invoke(quantize,
                               ["--model-dir", model_dir, "--dataset_dir", dataset_dir, "--output-dir", output_dir])

        # Verify that we got an error model directory not existing
        assert result.exit_code == 2
        assert "--model-dir" in result.output
        assert "Directory '{}' does not exist".format(model_dir) in result.output
    finally:
        if os.path.exists(tmp_dir):
            shutil.rmtree(tmp_dir)


@pytest.mark.common
def test_quantize_dataset_dir_does_not_exist():
    """
    Verifies that quantize command fails if the dataset directory does not exist
    """
    runner = CliRunner()

    tmp_dir = tempfile.mkdtemp()
    model_dir = os.path.join(tmp_dir, 'resnet_v1_50', '3')
    dataset_dir = os.path.join(tmp_dir, 'data')
    output_dir = os.path.join(tmp_dir, 'output')

    try:
        os.makedirs(model_dir)

        # Call the quantize command with the model directory
        result = runner.invoke(quantize,
                               ["--model-dir", model_dir, "--dataset_dir", dataset_dir, "--output-dir", output_dir])

        # Verify that we got an error dataset directory not existing
        assert result.exit_code == 2
        assert "--dataset-dir" in result.output
        assert "Directory '{}' does not exist".format(dataset_dir) in result.output
    finally:
        if os.path.exists(tmp_dir):
            shutil.rmtree(tmp_dir)


@pytest.mark.common
@patch("tlt.models.model_factory.get_model")
@patch("tlt.datasets.dataset_factory.load_dataset")
def test_quantize_output_dir(mock_get_model, mock_load_dataset):
    """
    Verifies that quantize command increments the output directory for the quantized model each time the quantization
    command is called
    """
    runner = CliRunner()

    tmp_dir = tempfile.mkdtemp()
    model_name = 'resnet_v1_50'
    model_dir = os.path.join(tmp_dir, model_name, '3')
    dataset_dir = os.path.join(tmp_dir, 'data')
    output_dir = os.path.join(tmp_dir, 'output')

    try:
        for new_dir in [model_dir, dataset_dir]:
            os.makedirs(new_dir)

        Path(os.path.join(model_dir, 'saved_model.pb')).touch()

        model_mock = MagicMock()
        data_mock = MagicMock()
        model_mock.use_case = "image_classification"
        data_mock.use_case = "image_classification"

        mock_get_model.return_value = model_mock
        mock_load_dataset.return_value = data_mock

        for i in range(1, 5):
            # Call the quantize command
            result = runner.invoke(quantize,
                                   ["--model-dir", model_dir, "--dataset_dir", dataset_dir,
                                    "--output-dir", output_dir])
            assert result.exit_code == 0

            # Check for an expected quantization output dir with the folder number incrementing
            expected_quantize_dir = os.path.join(output_dir, "quantize", model_name, str(i))
            model_mock.quantize.called_once_with(model_dir, expected_quantize_dir)

            model_mock.reset_mock()

    finally:
        if os.path.exists(tmp_dir):
            shutil.rmtree(tmp_dir)


class TestQuantizationArgs:
    """
    Class for tests that are testing bad inputs for quntization args with generic folders for the model dir,
    dataset dir, and output dir.
    """

    def setup_class(self):
        self._runner = CliRunner()

        self._tmp_dir = tempfile.mkdtemp()
        self._model_dir = os.path.join(self._tmp_dir, 'resnet_v1_50', '3')
        self._dataset_dir = os.path.join(self._tmp_dir, 'data')
        self._output_dir = os.path.join(self._tmp_dir, 'output')

    def setup_method(self):
        for new_dir in [self._model_dir, self._dataset_dir]:
            if not os.path.exists(new_dir):
                os.makedirs(new_dir)

    def teardown_method(self):
        if os.path.exists(self._tmp_dir):
            shutil.rmtree(self._tmp_dir)

    def teardown_class(self):
        if os.path.exists(self._tmp_dir):
            shutil.rmtree(self._tmp_dir)

    @pytest.mark.common
    @pytest.mark.parametrize('max_trials',
                             [-1, -5, 'foo'])
    def test_quantize_invalid_max_trials(self, max_trials):
        """
        Verifies that quantize command fails if the max trials is invalid (should be an integer > 0)
        """

        # Create the model file
        Path(os.path.join(self._model_dir, 'saved_model.pt')).touch()

        # Call the quantize command with the model directory
        result = self._runner.invoke(quantize,
                                     ["--model-dir", self._model_dir,
                                      "--dataset_dir", self._dataset_dir,
                                      "--output-dir", self._output_dir,
                                      "--max-trials", max_trials])

        assert result.exit_code == 2
        assert "Invalid value for '--max-trials'" in result.output

    @pytest.mark.common
    @pytest.mark.parametrize('timeout', [-1, -5, 'foo'])
    def test_quantize_invalid_timeout(self, timeout):
        """
        Verifies that quantize command fails if the timeout is invalid (should be an integer >= 0)
        """

        # Create the model file
        Path(os.path.join(self._model_dir, 'saved_model.pt')).touch()

        # Call the quantize command with the model directory
        result = self._runner.invoke(quantize,
                                     ["--model-dir", self._model_dir,
                                      "--dataset_dir", self._dataset_dir,
                                      "--output-dir", self._output_dir,
                                      "--timeout", timeout])

        assert result.exit_code == 2
        assert "Invalid value for '--timeout'" in result.output

    @pytest.mark.common
    @pytest.mark.parametrize('accuracy_criterion',
                             [1.3, -5, 'foo'])
    def test_quantize_invalid_accuracy_criterion(self, accuracy_criterion):
        """
        Verifies quantize command fails if the accuracy criterion value is invalid (should be a float between 0 and 1.0)
        """

        # Create the model file
        Path(os.path.join(self._model_dir, 'saved_model.pt')).touch()

        # Call the quantize command with the model directory
        result = self._runner.invoke(quantize,
                                     ["--model-dir", self._model_dir,
                                      "--dataset_dir", self._dataset_dir,
                                      "--output-dir", self._output_dir,
                                      "--accuracy-criterion", accuracy_criterion])

        assert result.exit_code == 2
        assert "Invalid value for '--accuracy-criterion'" in result.output