# Copyright 2017-2020 Amazon.com, Inc. or its affiliates. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License"). You
# may not use this file except in compliance with the License. A copy of
# the License is located at
#
#     http://aws.amazon.com/apache2.0/
#
# or in the "license" file accompanying this file. This file 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.
from __future__ import absolute_import

import os

import pytest
from sagemaker.huggingface import HuggingFace

from packaging.version import Version

from ..... import invoke_sm_helper_function
from ...integration.utils import processor, py_version, unique_name_from_base  # noqa: F401
from test.test_utils import get_framework_and_version_from_tag, get_cuda_version_from_tag

RESOURCE_PATH = os.path.join(os.path.dirname(__file__), "..", "..", "resources")
BERT_PATH = os.path.join(RESOURCE_PATH, "scripts")

# hyperparameters, which are passed into the training job
hyperparameters = {
    "max_steps": 10,
    "train_batch_size": 8,
    "model_name": "distilbert-base-uncased",
}


@pytest.mark.integration("hf_smdp")
@pytest.mark.model("hf_distilbert")
@pytest.mark.processor("gpu")
@pytest.mark.skip_cpu
@pytest.mark.skip_py2_containers
@pytest.mark.skip_trcomp_containers
# TODO: Enable sagemaker debugger, resolve github issue after enabling.
#  https://github.com/aws/deep-learning-containers/issues/1053
def test_hf_smdp(ecr_image, sagemaker_regions, instance_type, framework_version, tmpdir):
    """
    Tests SMDataParallel single-node command via script mode
    """
    invoke_sm_helper_function(
        ecr_image,
        sagemaker_regions,
        _test_hf_smdp_function,
        instance_type,
        framework_version,
        py_version,
        tmpdir,
        1,
    )


@pytest.mark.processor("gpu")
@pytest.mark.skip_cpu
@pytest.mark.multinode(2)
@pytest.mark.integration("hf_smdp_multinode")
@pytest.mark.model("hf_distilbert")
@pytest.mark.skip_py2_containers
@pytest.mark.skip_trcomp_containers
# Skipping `ml.p3dn.24xlarge` instance type due to capacity issue in us-west-2
# TODO: Enable sagemaker debugger, resolve github issue after enabling.
#  https://github.com/aws/deep-learning-containers/issues/1053
def test_hf_smdp_multi(ecr_image, sagemaker_regions, instance_type, tmpdir, framework_version):
    """
    Tests smddprun command via Estimator API distribution parameter
    """
    invoke_sm_helper_function(
        ecr_image,
        sagemaker_regions,
        _test_hf_smdp_function,
        instance_type,
        framework_version,
        py_version,
        tmpdir,
        2,
    )


def _test_hf_smdp_function(
    ecr_image,
    sagemaker_session,
    instance_type,
    framework_version,
    py_version,
    tmpdir,
    instance_count,
):
    _, image_framework_version = get_framework_and_version_from_tag(ecr_image)
    image_cuda_version = get_cuda_version_from_tag(ecr_image)

    instance_type = "ml.p3.16xlarge"
    distribution = {"smdistributed": {"dataparallel": {"enabled": True}}}

    estimator = HuggingFace(
        entry_point="train.py",
        source_dir=BERT_PATH,
        role="SageMakerRole",
        instance_type=instance_type,
        instance_count=instance_count,
        image_uri=ecr_image,
        framework_version=framework_version,
        py_version=py_version,
        sagemaker_session=sagemaker_session,
        hyperparameters=hyperparameters,
        distribution=distribution,
        debugger_hook_config=False,  # currently needed
    )

    estimator.fit(job_name=unique_name_from_base("test-tf-hf-smdp-multi"))