# Copyright (c) 2025, NVIDIA CORPORATION.  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.
# 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.

import nemo_run as run
import pytest

from nemo.collections.llm.api import finetune, pretrain
from nemo.collections.llm.gpt.data.mock import MockDataModule
from nemo.collections.llm.gpt.data.squad import SquadDataModule
from nemo.collections.llm.gpt.model.mistral import MistralConfig7B, MistralModel
from nemo.collections.llm.peft.lora import LoRA
from nemo.collections.llm.recipes import mistral_7b as mistral
from nemo.lightning import Trainer


class TestMistral:
    @pytest.fixture(scope="class")
    def recipe_module(self):
        return mistral

    def test_model(self, recipe_module):
        model_config = recipe_module.model()
        assert isinstance(model_config, run.Config)
        assert model_config.__fn_or_cls__ == MistralModel
        assert isinstance(model_config.config, run.Config)
        assert model_config.config.__fn_or_cls__ == MistralConfig7B

    def test_trainer(self, recipe_module):
        trainer_config = recipe_module.trainer()
        assert isinstance(trainer_config, run.Config)
        assert trainer_config.__fn_or_cls__ == Trainer
        assert trainer_config.accelerator == "gpu"
        assert trainer_config.devices == 8
        assert trainer_config.num_nodes == 1

        # Check strategy configuration
        assert isinstance(trainer_config.strategy, run.Config)
        assert trainer_config.strategy.__fn_or_cls__.__name__ == "MegatronStrategy"
        assert trainer_config.strategy.tensor_model_parallel_size == 1
        assert trainer_config.strategy.pipeline_model_parallel_size == 1
        assert trainer_config.strategy.pipeline_dtype is None
        assert trainer_config.strategy.virtual_pipeline_model_parallel_size is None
        assert trainer_config.strategy.context_parallel_size == 2
        assert trainer_config.strategy.sequence_parallel is False

    def test_pretrain_recipe(self, recipe_module):
        recipe = recipe_module.pretrain_recipe()
        assert isinstance(recipe, run.Partial)
        assert recipe.__fn_or_cls__ == pretrain
        assert isinstance(recipe.model, run.Config)
        assert recipe.model.__fn_or_cls__ == MistralModel
        assert isinstance(recipe.trainer, run.Config)
        assert recipe.trainer.__fn_or_cls__ == Trainer
        assert isinstance(recipe.data, run.Config)
        assert recipe.data.__fn_or_cls__ == MockDataModule
        assert recipe.data.seq_length == 4096
        assert recipe.data.global_batch_size == 512
        assert recipe.data.micro_batch_size == 1

    def test_finetune_recipe(self, recipe_module):
        recipe = recipe_module.finetune_recipe()
        assert isinstance(recipe, run.Partial)
        assert recipe.__fn_or_cls__ == finetune
        assert isinstance(recipe.model, run.Config)
        assert recipe.model.__fn_or_cls__ == MistralModel
        assert isinstance(recipe.trainer, run.Config)
        assert recipe.trainer.__fn_or_cls__ == Trainer
        assert isinstance(recipe.data, run.Config)
        assert recipe.data.__fn_or_cls__ == SquadDataModule
        assert recipe.data.seq_length == 2048
        assert recipe.data.global_batch_size == 128
        assert recipe.data.micro_batch_size == 1
        assert isinstance(recipe.peft, run.Config)
        assert recipe.peft.__fn_or_cls__ == LoRA

    @pytest.mark.parametrize("num_nodes,num_gpus_per_node", [(1, 8), (2, 4), (4, 2)])
    def test_pretrain_recipe_with_different_configurations(self, recipe_module, num_nodes, num_gpus_per_node):
        recipe = recipe_module.pretrain_recipe(num_nodes=num_nodes, num_gpus_per_node=num_gpus_per_node)
        assert recipe.trainer.num_nodes == num_nodes
        assert recipe.trainer.devices == num_gpus_per_node

    def test_trainer_parallelism_options(self, recipe_module):
        trainer_config = recipe_module.trainer(
            tensor_parallelism=2, pipeline_parallelism=2, context_parallelism=4, sequence_parallelism=True
        )
        assert trainer_config.strategy.tensor_model_parallel_size == 2
        assert trainer_config.strategy.pipeline_model_parallel_size == 2
        assert trainer_config.strategy.context_parallel_size == 4
        assert trainer_config.strategy.sequence_parallel is True

    def test_model_config_parameters(self, recipe_module):
        model_config = recipe_module.model()
        mistral_config = model_config.config
        assert mistral_config.num_layers == 32
        assert mistral_config.hidden_size == 4096
        assert mistral_config.num_attention_heads == 32
        assert mistral_config.seq_length == 32768