# Adopted from tatsu-lab@stanford_alpaca. Below is the original copyright:
#    Copyright 2023 Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li
#
#    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 copy
from dataclasses import dataclass, field
import json
import logging
import pathlib
from typing import Dict, Optional, Sequence

import torch

import transformers
from torch.utils.data import Dataset
from transformers import Trainer

from fastchat import conversation as conversation_lib

# TODO: import and use code from ../data/dataset.py

IGNORE_INDEX = -100
DEFAULT_PAD_TOKEN = "[PAD]"
DEFAULT_EOS_TOKEN = "</s>"
DEFAULT_BOS_TOKEN = "</s>"
DEFAULT_UNK_TOKEN = "</s>"


@dataclass
class ModelArguments:
    model_name_or_path: Optional[str] = field(default="facebook/opt-125m")


@dataclass
class DataArguments:
    data_path: str = field(default=None,
                           metadata={"help": "Path to the training data."})
    lazy_preprocess: bool = False


@dataclass
class TrainingArguments(transformers.TrainingArguments):
    cache_dir: Optional[str] = field(default=None)
    optim: str = field(default="adamw_torch")
    model_max_length: int = field(
        default=512,
        metadata={
            "help":
            "Maximum sequence length. Sequences will be right padded (and possibly truncated)."
        },
    )


def safe_save_model_for_hf_trainer(trainer: transformers.Trainer,
                                   output_dir: str):
    """Collects the state dict and dump to disk."""
    state_dict = trainer.model.state_dict()
    if trainer.args.should_save:
        cpu_state_dict = {
            key: value.cpu()
            for key, value in state_dict.items()
        }
        del state_dict
        trainer._save(output_dir, state_dict=cpu_state_dict)  # noqa


def smart_tokenizer_and_embedding_resize(
    special_tokens_dict: Dict,
    tokenizer: transformers.PreTrainedTokenizer,
    model: transformers.PreTrainedModel,
):
    """Resize tokenizer and embedding.

    Note: This is the unoptimized version that may make your embedding size not be divisible by 64.
    """
    num_new_tokens = tokenizer.add_special_tokens(special_tokens_dict)
    model.resize_token_embeddings(len(tokenizer))

    if num_new_tokens > 0:
        input_embeddings = model.get_input_embeddings().weight.data
        output_embeddings = model.get_output_embeddings().weight.data

        input_embeddings_avg = input_embeddings[:-num_new_tokens].mean(
            dim=0, keepdim=True)
        output_embeddings_avg = output_embeddings[:-num_new_tokens].mean(
            dim=0, keepdim=True)

        input_embeddings[-num_new_tokens:] = input_embeddings_avg
        output_embeddings[-num_new_tokens:] = output_embeddings_avg


def _tokenize_fn(strings: Sequence[str],
                 tokenizer: transformers.PreTrainedTokenizer) -> Dict:
    """Tokenize a list of strings."""
    tokenized_list = [
        tokenizer(
            text,
            return_tensors="pt",
            padding="longest",
            max_length=tokenizer.model_max_length,
            truncation=True,
        ) for text in strings
    ]
    input_ids = labels = [
        tokenized.input_ids[0] for tokenized in tokenized_list
    ]
    input_ids_lens = labels_lens = [
        tokenized.input_ids.ne(tokenizer.pad_token_id).sum().item()
        for tokenized in tokenized_list
    ]
    return dict(
        input_ids=input_ids,
        labels=labels,
        input_ids_lens=input_ids_lens,
        labels_lens=labels_lens,
    )


def _mask_targets(target, tokenized_lens, speakers, header_len, s_ids):
    cur_idx = header_len
    tgt_len = target.shape[0]
    for tokenized_len, speaker, s_id in zip(tokenized_lens, speakers, s_ids):
        if cur_idx >= tgt_len:
            break
        elif cur_idx + tokenized_len < tgt_len:
            # Check whether the mask is applied to the correct position
            if not torch.equal(target[cur_idx + 2:cur_idx + tokenized_len],
                               s_id[2:]):
                logging.warning("a sentence mismatches the corresponding piece "
                                "in the conversation")
        if speaker == "human":
            target[cur_idx:cur_idx + tokenized_len] = IGNORE_INDEX
        cur_idx += tokenized_len


def _add_speaker_and_signal(header, source, get_conversation=True):
    """Add speaker and start/end signal on each round."""
    BEGIN_SIGNAL = "### "
    END_SIGNAL = "\n"
    conversation = header
    unknown_role = "unknown"  # use default unknown role
    roles = {
        "human": conversation_lib.default_conversation.roles[0],  # human role
        "gpt": conversation_lib.default_conversation.roles[1],  # gpt role
    }
    for sentence in source:
        sentence_from = sentence["from"].lower()
        sentence["value"] = (
            BEGIN_SIGNAL
            + roles.get(sentence_from, unknown_role)
            + ": "
            + sentence["value"]
            + END_SIGNAL
        )
        if get_conversation:
            conversation += sentence["value"]
    return conversation


def preprocess(
    sources: Sequence[str],
    tokenizer: transformers.PreTrainedTokenizer,
) -> Dict:
    """
    Given a list of sources, each is a conversation list. This transform:
    1. Add signal '### ' at the beginning each sentence, with end signal '\n';
    2. Concatenate conversations together;
    3. Tokenize the concatenated conversation;
    4. Make a deepcopy as the target. Mask human words with IGNORE_INDEX.
    """
    # add end signal and concatenate together
    conversations = []
    header = f"{conversation_lib.default_conversation.system}\n\n"
    for source in sources:
        conversation = _add_speaker_and_signal(header, source)
        conversations.append(conversation)
    # tokenize conversations
    conversations_tokenized = _tokenize_fn(conversations, tokenizer)
    input_ids = conversations_tokenized["input_ids"]
    targets = copy.deepcopy(input_ids)
    header_len = _tokenize_fn([header], tokenizer)["input_ids_lens"][0]
    for target, source in zip(targets, sources):
        tokenized_sentence = _tokenize_fn([s["value"] for s in source], tokenizer)
        tokenized_lens = tokenized_sentence["input_ids_lens"]
        # Currently, "###" is tokenized into 2 tokens in the whole conversation,
        # and 1 token in a single sentence, so we do not need to use the line below.
        # tokenized_lens = [l-1 for l in tokenized_lens]
        speakers = [sentence["from"] for sentence in source]
        ids = tokenized_sentence["input_ids"]
        _mask_targets(target, tokenized_lens, speakers, header_len, ids)

    return dict(input_ids=input_ids, labels=targets)


class SupervisedDataset(Dataset):
    """Dataset for supervised fine-tuning."""

    def __init__(self, data_path: str,
                 tokenizer: transformers.PreTrainedTokenizer):
        super(SupervisedDataset, self).__init__()
        logging.warning("Loading data...")
        list_data_dict = json.load(open(data_path, "r"))

        logging.warning("Formatting inputs...")
        sources = [example["conversations"] for example in list_data_dict]
        data_dict = preprocess(sources, tokenizer)

        self.input_ids = data_dict["input_ids"]
        self.labels = data_dict["labels"]

    def __len__(self):
        return len(self.input_ids)

    def __getitem__(self, i) -> Dict[str, torch.Tensor]:
        return dict(input_ids=self.input_ids[i], labels=self.labels[i])


class LazySupervisedDataset(Dataset):
    """Dataset for supervised fine-tuning."""

    def __init__(self, data_path: str,
                 tokenizer: transformers.PreTrainedTokenizer):
        super(LazySupervisedDataset, self).__init__()
        logging.warning("Loading data...")
        list_data_dict = json.load(open(data_path, "r"))

        logging.warning("Formatting inputs...Skip in lazy mode")
        self.tokenizer = tokenizer
        self.list_data_dict = list_data_dict

    def __len__(self):
        return len(self.list_data_dict)

    def __getitem__(self, i) -> Dict[str, torch.Tensor]:
        sources = self.list_data_dict[i]
        if isinstance(i, int):
            sources = [sources]
        data_dict = preprocess(
            copy.deepcopy([e["conversations"] for e in sources]),
            self.tokenizer)
        if isinstance(i, int):
            data_dict = dict(input_ids=data_dict["input_ids"][0],
                             labels=data_dict["labels"][0])
        return data_dict


@dataclass
class DataCollatorForSupervisedDataset(object):
    """Collate examples for supervised fine-tuning."""

    tokenizer: transformers.PreTrainedTokenizer

    def __call__(self, instances: Sequence[Dict]) -> Dict[str, torch.Tensor]:
        input_ids, labels = tuple([instance[key] for instance in instances]
                                  for key in ("input_ids", "labels"))
        input_ids = torch.nn.utils.rnn.pad_sequence(
            input_ids,
            batch_first=True,
            padding_value=self.tokenizer.pad_token_id)
        labels = torch.nn.utils.rnn.pad_sequence(labels,
                                                 batch_first=True,
                                                 padding_value=IGNORE_INDEX)
        return dict(
            input_ids=input_ids,
            labels=labels,
            attention_mask=input_ids.ne(self.tokenizer.pad_token_id),
        )


def make_supervised_data_module(tokenizer: transformers.PreTrainedTokenizer,
                                data_args) -> Dict:
    """Make dataset and collator for supervised fine-tuning."""
    dataset_cls = (LazySupervisedDataset
                   if data_args.lazy_preprocess else SupervisedDataset)
    train_dataset = dataset_cls(tokenizer=tokenizer,
                                data_path=data_args.data_path)
    data_collator = DataCollatorForSupervisedDataset(tokenizer=tokenizer)
    return dict(train_dataset=train_dataset,
                eval_dataset=None,
                data_collator=data_collator)


def train():
    parser = transformers.HfArgumentParser(
        (ModelArguments, DataArguments, TrainingArguments))
    model_args, data_args, training_args = parser.parse_args_into_dataclasses()
    model = transformers.LlamaForCausalLM.from_pretrained(
        model_args.model_name_or_path,
        cache_dir=training_args.cache_dir,
    )

    tokenizer = transformers.AutoTokenizer.from_pretrained(
        model_args.model_name_or_path,
        cache_dir=training_args.cache_dir,
        model_max_length=training_args.model_max_length,
        padding_side="right",
        use_fast=False,
    )
    if tokenizer.pad_token is None:
        smart_tokenizer_and_embedding_resize(
            special_tokens_dict=dict(pad_token=DEFAULT_PAD_TOKEN),
            tokenizer=tokenizer,
            model=model,
        )
    if "llama" in model_args.model_name_or_path:
        tokenizer.add_special_tokens({
            "eos_token": DEFAULT_EOS_TOKEN,
            "bos_token": DEFAULT_BOS_TOKEN,
            "unk_token": DEFAULT_UNK_TOKEN,
        })

    data_module = make_supervised_data_module(tokenizer=tokenizer,
                                              data_args=data_args)
    trainer = Trainer(model=model,
                    tokenizer=tokenizer,
                    args=training_args,
                    **data_module)

    if list(pathlib.Path(training_args.output_dir).glob("checkpoint-*")):
        trainer.train(resume_from_checkpoint=True)
    else:
        trainer.train()
    trainer.save_state()
    safe_save_model_for_hf_trainer(trainer=trainer,
                                   output_dir=training_args.output_dir)


if __name__ == "__main__":
    train()