docker/intel_code/llama13b/Model-References/modifytrain.py

from datasets import concatenate_datasets, load_dataset, load_from_disk
import argparse
from tokenizers import Tokenizer, decoders, models, pre_tokenizers, processors, trainers
from transformers import GPT2TokenizerFast, AutoTokenizer
from datasets import config
import logging

def initialize_logger(log_file):
    logging.basicConfig(filename=log_file, level=logging.INFO, format='%(asctime)s: %(message)s')

def log_parameters(vocab_size, batch_size, fertility_score, proportion_continued_words, log_file='parameters.log'):
    initialize_logger(log_file)
    logging.info(f"Vocabulary Size: {vocab_size}, Batch Size: {batch_size}, Fertility Score: {fertility_score}, Proportion of Continued word: {proportion_continued_words}")

def parse_arguments():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--batch_size",
        type=int,
        required=True,
        help="Batch size to use for training"
    )
    parser.add_argument(
        "--vocab_size",
        type=int,
        required=True,
        help="Vocabulary size to use for tokenizer"
    )
    parser.add_argument(
        "--use_config",
        choices=['xlm-roberta', 'vanilla'],
        required=True,
        help="Use XLM-RoBERTa config or Vanilla BPE"
    )
    parser.add_argument(
        "--do_evaluate",
        action='store_true',
        help="Enable evaluation."
    )
    args = parser.parse_known_args()
    return args

def calculate_proportion_continued_words(tokenizer, sentences):
    total_continued_words = 0
    total_words = 0
    for sentence in sentences:
        tok = tokenizer.encode_plus(sentence, return_tensors="pt")
        input_ids = tok['input_ids'].squeeze(0)
        continued_word = False
        for i in range(1, len(input_ids)):
            if input_ids[i] != tokenizer.pad_token_id:
                if continued_word:
                    total_continued_words += 1
                continued_word = True
            else:
                continued_word = False
        total_words += len(sentence.split())
    proportion_continued_words = total_continued_words / total_words if total_words > 0 else 0
    return proportion_continued_words



def train_tokenizer(args):
    
    # configs = ['as', 'bd', 'bn', 'dg', 'en', 'gom', 'gu', 'hi', 'kha', 'kn', 'ks', 'mai', 'ml', 'mni', 'mr', 'ne', 'or', 'pa', 'sa', 'sat', 'sd', 'ta', 'te', 'ur']
    indic_datasets = []
    configs=['hi']
    # for c in configs:
        # indic_dataset = load_dataset('satpalsr/indicCorpv2', c, split='train', cache_dir='/home1/BharatGPT_tokenizer/hf/')
        # indic_datasets.extend(indic_dataset)
    # wikidataset= load_dataset('wiki40b', 'en', split='train', cache_dir='/home1/BharatGPT_tokenizer/hf/')
    indic_datasets_hi= load_dataset('satpalsr/indicCorpv2', 'hi', split='train', cache_dir='/home1/BharatGPT_tokenizer/hf/')
    indic_datasets_en= load_dataset('satpalsr/indicCorpv2', 'en', split='train', cache_dir='/home1/BharatGPT_tokenizer/hf/')
    
    #  = wikidataset.remove_columns(['wikidata_id', 'version_id'])
    print(indic_datasets)
    # print(wikidataset)
    dataset = concatenate_datasets([indic_datasets_en,indic_datasets_hi])
    #test_data = load_from_disk('samanantar_data')
    #test_data = dataset['text'][:10000]
    #print(f"[INFO] {len(test_data)}")
    #print(f"[INFO] {len(dataset)}")
    # print(f"[INFO] {test_data[:10]}")

    def batch_iterator():
        for idx in range(0, len(dataset), args.batch_size):
            yield dataset[idx: idx + args.batch_size]['text']

    if args.use_config == 'vanilla':
        tokenizer = Tokenizer(models.BPE())
        tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(add_prefix_space=False)
        print(f"[INFO] The brown fox jumped over the lazy dog\n{tokenizer.pre_tokenizer.pre_tokenize_str('The brown fox jumped over the lazy dog')}")
        print(f"[INFO] Training...")
        trainer = trainers.BpeTrainer(vocab_size=args.vocab_size, special_tokens=["<|endoftext|>"])
        tokenizer.train_from_iterator(batch_iterator(), trainer=trainer)
        tokenizer.post_processor = processors.ByteLevel(trim_offsets=False)
        tokenizer.decoder = decoders.ByteLevel()
        trained_tokenizer = GPT2TokenizerFast(tokenizer_object=tokenizer)
    elif args.use_config == 'xlm-roberta':
        print("skipped")
        tokenizer = AutoTokenizer.from_pretrained('xlm-roberta-base')
        trained_tokenizer = tokenizer.train_new_from_iterator(batch_iterator(), vocab_size=args.vocab_size)

    trained_tokenizer.save_pretrained('hi-bgpt-bpe-tokenizer1')
    print(f"[INFO] Tokenizer saved to disk")

    if args.do_evaluate:
        print(f"[INFO] Running evaluation using fertility and fraction of continued words")
        tokenizer = AutoTokenizer.from_pretrained('hi-bgpt-bpe-tokenizer1')
        # tokenizer = AutoTokenizer.from_pretrained('xlm-roberta-base')
        fertility = 0
        for sentence in test_data: 
            tok=tokenizer(sentence)
            fertility += len(tok['input_ids']) / len(sentence.split())
        average_fertility = fertility / len(test_data)
        proportion_continued_words = calculate_proportion_continued_words(tokenizer, test_data)
        log_parameters(args.vocab_size, args.batch_size, average_fertility, proportion_continued_words)

def main():
    args, _ = parse_arguments()
    train_tokenizer(args)

if __name__ == "__main__":
    main()