# 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.


from typing import Optional

import lightning.pytorch as pl
import nemo_run as run
import torch
from megatron.core.distributed import DistributedDataParallelConfig

import nemo.lightning as nl
from nemo.collections.llm.recipes.precision.mixed_precision import bf16_mixed


def trainer(
    tensor_parallelism: int = 1,
    pipeline_parallelism: int = 1,
    pipeline_parallelism_type: Optional[torch.dtype] = torch.bfloat16,
    virtual_pipeline_parallelism: Optional[int] = None,
    context_parallelism: int = 1,
    expert_parallelism: int = 1,
    sequence_parallelism: bool = True,
    num_nodes: int = 1,
    num_gpus_per_node: int = 8,
    max_steps: int = 1168251,
    callbacks: Optional[list[run.Config[pl.Callback]]] = None,
) -> run.Config[nl.Trainer]:
    """
    Configure the NeMo Lightning Trainer for DeepSeek V2/V3 models.

    This function sets up the distributed training strategy optimized for DeepSeek.

    Args:
        tensor_parallelism (int): Degree of tensor model parallelism.
        pipeline_parallelism (int): Degree of pipeline model parallelism.
        pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism.
        virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism.
        context_parallelism (int): Degree of context parallelism.
        expert_parallelism (int): Degree of expert parallelism.
        sequence_parallelism (bool): Whether to use sequence parallelism.
        num_nodes (int): Number of compute nodes to use.
        num_gpus_per_node (int): Number of GPUs per node.
        max_steps (int): Maximum number of training steps.
        callbacks (Optional[list[run.Config[Callback]]]): List of callback configurations.

    Returns:
        run.Config[nl.Trainer]: Configuration for the NeMo Lightning Trainer.

    """
    strategy = run.Config(
        nl.MegatronStrategy,
        tensor_model_parallel_size=tensor_parallelism,
        pipeline_model_parallel_size=pipeline_parallelism,
        pipeline_dtype=pipeline_parallelism_type,
        virtual_pipeline_model_parallel_size=virtual_pipeline_parallelism,
        context_parallel_size=context_parallelism,
        expert_model_parallel_size=expert_parallelism,
        sequence_parallel=sequence_parallelism,
        gradient_as_bucket_view=True,
        ckpt_async_save=True,
        ckpt_parallel_load=True,
        ddp=run.Config(
            DistributedDataParallelConfig,
            check_for_nan_in_grad=True,
            grad_reduce_in_fp32=True,
            overlap_grad_reduce=True,
            overlap_param_gather=True,
            average_in_collective=True,
        ),
    )

    trainer = run.Config(
        nl.Trainer,
        accelerator="gpu",
        accumulate_grad_batches=1,
        callbacks=callbacks,
        devices=num_gpus_per_node,
        limit_test_batches=50,
        limit_val_batches=32,
        log_every_n_steps=10,
        max_steps=max_steps,
        num_nodes=num_nodes,
        plugins=bf16_mixed(),
        strategy=strategy,
        use_distributed_sampler=False,
        val_check_interval=2000,
        num_sanity_val_steps=0,
    )

    return trainer