from typing import Optional
from functools import partial

from datasets import load_dataset
from litdata import optimize, TokensLoader
from litgpt.tokenizer import Tokenizer


def batch_iterator(path: str,
                   name: Optional[str]=None,
                   data_dir: Optional[str]=None,
                   data_files: Optional[str]=None,
                   revision: Optional[str]=None,
                   split: str='train',
                   format: Optional[str]=None):
    assert format is not None
    
    dataset = load_dataset(path=path,
                           name=name,
                           data_dir=data_dir,
                           data_files=data_files,
                           revision=revision,
                           split=split,
                           trust_remote_code=True)

    for row in dataset:
        text = format.format(**row)
        yield text


def tokenize_fn(datasets_config, tokenizer=None):
    for text in batch_iterator(**datasets_config):
        text_ids = tokenizer.encode(text, bos=False, eos=True)
        yield text_ids


datasets_configs = [
    {'path': 'yahma/alpaca-cleaned', 'format': '{instruction} {input} {output}'},
    {'path': 'gbharti/wealth-alpaca_lora', 'format': '{instruction} {input} {output}'},
    *[
        {'path': 'saillab/taco-datasets', 'data_dir': data_dir, 'split': 'train[:10%]', 'format': '{instruction} {input} {output}'}
        for data_dir in [
            'multilingual-instruction-tuning-dataset /multilingual-alpaca-52k-gpt-4',
            'multilingual-instruction-tuning-dataset /multilinugal-dolly-15k',
        ]
    ],
    *[
        {'path': 'xu-song/cc100-samples', 'name': name, 'split': 'train[:10%]', 'format': '{text}'}
        for name in [
            'am', 'ar', 'as', 'az', 'be', 'bg', 'bn', 'bn_rom', 'br',
            'bs', 'ca', 'cs', 'cy', 'da', 'de', 'el', 'en', 'eo', 'es',
            'et', 'eu', 'fa', 'ff', 'fi', 'fr', 'fy', 'ga', 'gd', 'gl',
            'gn', 'gu', 'ha', 'he', 'hi', 'hi_rom', 'hr', 'ht', 'hu',
            'hy', 'id', 'ig', 'is', 'it', 'ja', 'jv', 'ka', 'kk', 'km',
            'kn', 'ko', 'ku', 'ky', 'la', 'lg', 'li', 'ln', 'lo', 'lt',
            'lv', 'mg', 'mk', 'ml', 'mn', 'mr', 'ms', 'my', 'my_zaw',
            'ne', 'nl', 'no', 'ns', 'om', 'or', 'pa', 'pl', 'ps', 'pt',
            'qu', 'rm', 'ro', 'ru', 'sa', 'si', 'sc', 'sd', 'sk', 'sl',
            'so', 'sq', 'sr', 'ss', 'su', 'sv', 'sw', 'ta', 'ta_rom',
            'te', 'te_rom', 'th', 'tl', 'tn', 'tr', 'ug', 'uk', 'ur',
            'ur_rom', 'uz', 'vi', 'wo', 'xh', 'yi', 'yo',
            'zh-Hans', 'zh-Hant', 'zu',
        ]
    ],
    {'path': 'ontocord/fineweb-permissive-multilingual-2m', 'split': 'train[:5%]', 'format': '{text}'},
    {'path': 'MuskumPillerum/General-Knowledge', 'format': '{Question} {Answer}'},
    {'path': 'yirenc/general_knowledge_boolean', 'split': 'train+validation', 'format': '{question}? {answer}. {passage}'},
    {'path': 'nampdn-ai/tiny-textbooks', 'split': 'train+test', 'format': '{textbook}'},
    {'path': 'nampdn-ai/tiny-codes', 'split': 'train[:5%]', 'format': '{prompt} {response}'},
    *[
        {'path': 'bigcode/the-stack-smol-xs', 'name': name, 'format': '{content}'}
        for name in [
            'ada', 'agda', 'alloy', 'antlr', 'applescript', 'assembly',
            'augeas', 'awk', 'batchfile', 'bison', 'bluespec', 'c',
            'c++', 'c-sharp', 'clojure', 'cmake', 'coffeescript', 'common-lisp',
            'css', 'cuda', 'dart', 'dockerfile', 'elixir',
            'elm', 'emacs-lisp','erlang', 'f-sharp', 'fortran', 'glsl', 'go',
            'groovy', 'haskell','html', 'idris', 'isabelle', 'java', 
            'java-server-pages', 'javascript', 'julia', 'kotlin', 'lean',
            'literate-agda', 'literate-coffeescript', 'literate-haskell',
            'lua', 'makefile', 'maple', 'markdown', 'mathematica', 'matlab',
            'ocaml', 'pascal', 'perl', 'php', 'powershell', 'prolog',
            'protocol-buffer', 'python', 'r', 'racket', 'restructuredtext',
            'rmarkdown', 'ruby', 'rust', 'sas', 'scala', 'scheme', 
            'shell', 'smalltalk', 'solidity', 'sparql', 'sql', 'stan',
            'standard-ml', 'stata', 'systemverilog', 'tcl', 'tcsh', 'tex', 
            'thrift', 'typescript', 'verilog', 'vhdl', 'visual-basic', 'xslt',
            'yacc', 'zig',
        ]
    ],
    {'path': 'm-a-p/CodeFeedback-Filtered-Instruction', 'split': 'train', 'format': '{query} {answer}'},
    {'path': 'jtatman/python-code-dataset-500k', 'format': '{instruction} {output}'},
    {'path': 'iamtarun/python_code_instructions_18k_alpaca', 'format': '{instruction} {input} {output}'},
    {'path': 'HuggingFaceH4/CodeAlpaca_20K', 'split': 'train+test', 'format': '{prompt} {completion}'},
    {'path': 'gair-prox/open-web-math-pro', 'split': 'train[:5%]', 'format': '{text}'},
    {'path': 'rvv-karma/Math-QA', 'split': 'train+val+test', 'format': '{question} {answer}'},
    {'path': 'ajibawa-2023/Maths-College', 'split': 'train[:10%]', 'format': '{instruction} {output}'},
    {'path': 'microsoft/orca-math-word-problems-200k', 'format': '{question} {answer}'},
    {'path': 'fblgit/simple-math', 'revision': 'refs/convert/parquet', 'split': 'train+test', 'format': '{instruction} = {output}'},
    {'path': 'SkunkworksAI/reasoning-0.01', 'format': '{instruction} {reasoning} {output}'},
    {'path': 'badrex/llm-emoji-dataset', 'format': '{character} {unicode} {short description} {tags} {LLM description}'},
]

outputs = optimize(
    fn=partial(tokenize_fn, tokenizer=Tokenizer('..')),
    inputs=datasets_configs,
    output_dir='../pretrain-data/',
    # Number of tokens to store by chunks. This is roughly 64MB of tokens per chunk.
    chunk_size=(2049 * 8012),
    num_workers=32,
)