# Copyright 2024 Huawei Technologies Co., Ltd
#
# 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.
# ============================================================================

"""
transform huggingface model to mindspore ckpt.
"""

import argparse
import os
from glob import glob
from transformers import LlamaForCausalLM
from safetensors.torch import load_file
import torch

import mindspore as ms

from mindformers.utils.convert_utils import pt2ms



def name_replace(weight_name: str):
    """replace weight name"""
    weight_name = weight_name.replace('embed_tokens.', 'tok_embeddings.')
    weight_name = weight_name.replace('lm_head.', 'output.')
    weight_name = weight_name.replace('.self_attn.q_proj.', '.attention.wq.')
    weight_name = weight_name.replace('.self_attn.k_proj.', '.attention.wk.')
    weight_name = weight_name.replace('.self_attn.v_proj.', '.attention.wv.')
    weight_name = weight_name.replace('.self_attn.o_proj.', '.attention.wo.')
    weight_name = weight_name.replace('.mlp.gate_proj.', '.feed_forward.w1.')
    weight_name = weight_name.replace('.mlp.down_proj.', '.feed_forward.w2.')
    weight_name = weight_name.replace('.mlp.up_proj.', '.feed_forward.w3.')
    weight_name = weight_name.replace('.input_layernorm.', '.attention_norm.')
    weight_name = weight_name.replace('.post_attention_layernorm.', '.ffn_norm.')
    return weight_name


def merge_safetensors_to_bin(input_path, file_path_):
    """
    merge safetensors to bin
    """
    merged_state_dict = {}

    safetensors = glob(os.path.join(input_path, '*.safetensors'))
    for safetensor in safetensors:
        # 加载safetensors文件
        print(f"===============Load file:{safetensor}")
        state_dict = load_file(safetensor)
        # 将state_dict 转换为torch的state_dict格式
        torch_state_dict = {key: value.clone().detach() for key, value in state_dict.items()}
        # 合并到merged_state_dict
        print(f"===============Save weight:{safetensor}")
        merged_state_dict.update(torch_state_dict)

    #临时文件，保存为bin格式
    if not os.path.exists('temp'):
        os.mkdir('temp')
    torch.save(merged_state_dict, file_path_)


# pylint: disable=W0613
def convert_pt_to_ms(input_path, output_path, dtype=None, **kwargs):
    """
    convert pt tp ms
    """

    file_path = os.path.join(input_path, "pytorch_model.bin")
    merge_safetensors_to_bin(input_path, file_path)

    print(f"Trying to convert mindspore checkpoint in {input_path}.")
    model_hf = LlamaForCausalLM.from_pretrained(os.path.dirname(file_path))
    if os.path.exists(file_path):
        os.remove(file_path)
    ckpt_list = []
    for name, value in model_hf.state_dict().items():
        name = name_replace(name)
        if name == 'model.norm.weight':
            name = 'model.norm_out.weight'
        if name == 'output.weight':
            name = 'lm_head.weight'
        if name == 'model.tok_embeddings.weight':
            name = 'model.tok_embeddings.embedding_weight'
        value = pt2ms(value, dtype)
        print(name, value.shape)
        ckpt_list.append({'name': name, 'data': value})

    ms.save_checkpoint(ckpt_list, output_path)
    print(f"Convert finished, the output is saved to {output_path}.")


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument('--torch_ckpt_path', default='./hf.bin')
    parser.add_argument('--mindspore_ckpt_path', default='transform.ckpt')
    args = parser.parse_args()
    convert_pt_to_ms(args.torch_ckpt_path, args.mindspore_ckpt_path)