Module molcrawl.genome_sequence.dataset.prepare_gpt2

Functions

def concatenate_texts(examples, eos_token_id)

Expand source code

def concatenate_texts(examples, eos_token_id):
    concatenated_ids = []
    for input_ids in examples["input_ids"]:
        concatenated_ids.extend(input_ids + [eos_token_id])
    return {"input_ids": concatenated_ids}

def create_chunks(examples, context_length)

Expand source code

def create_chunks(examples, context_length):
    concatenated_ids = examples["input_ids"]

    # Calculate the total number of chunks
    total_length = len(concatenated_ids)
    num_chunks = total_length // context_length

    # Truncate the concatenated_ids to a multiple of context_length
    total_length = num_chunks * context_length
    concatenated_ids = concatenated_ids[:total_length]

    # Split into chunks
    input_ids = [concatenated_ids[i : i + context_length] for i in range(0, total_length, context_length)]

    return {"input_ids": input_ids}

def tokenize_batch_dataset(output_dir, context_length, number_sample)

Expand source code

def tokenize_batch_dataset(output_dir, context_length, number_sample):
    from datasets import DatasetDict, load_dataset
    import sentencepiece as spm

    data = (
        load_dataset(
            "parquet",
            data_files=[str(Path(output_dir) / "parquet_files")],
            cache_dir=str(Path(output_dir) / "hf_cache"),
            split="train",
        )
        .shuffle()
        .select(range(number_sample))
    )

    tokenized_datasets = data.train_test_split(test_size=0.2)
    valid_test_split = tokenized_datasets["test"].train_test_split(test_size=0.5)
    tokenized_datasets = DatasetDict(
        {"train": tokenized_datasets["train"], "valid": valid_test_split["train"], "test": valid_test_split["test"]}
    )

    tokenizer = spm.SentencePieceProcessor(model_file=str(Path(output_dir) / "spm_tokenizer.model"))
    # tokenizer = AutoTokenizer.from_pretrained("zhihan1996/DNABERT-2-117M", trust_remote_code=True)

    concatenated_dataset = tokenized_datasets.map(
        partial(concatenate_texts, eos_token_id=tokenizer.eos_id()),
        # partial(concatenate_texts, eos_token_id=tokenizer.eos_token_id),
        batched=True,
        batch_size=-1,
        remove_columns=["num_tokens"],
    )

    chunked_dataset = concatenated_dataset.map(
        partial(create_chunks, context_length=context_length), batched=True, batch_size=context_length * 10
    )

    path_dataset = str(Path(output_dir) / "training_ready_hf_dataset")
    print(f"Saving dataset to: {path_dataset}. Match this path to the train_gpt2_config.py->dataset_dir parameter.")
    chunked_dataset.save_to_disk(path_dataset)

def tokenize_function(examples, tokenizer)

Expand source code

def tokenize_function(examples, tokenizer):
    return {"input_ids": tokenizer.encode(examples["text"]).ids}