isoformer_tokenizer.py

# coding=utf-8
# Copyright 2022 Meta and The HuggingFace Inc. team. All rights reserved.
#
# 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.
"""Tokenization classes for ESM."""
import os
from huggingface_hub import hf_hub_download
from typing import List, Optional

#from transformers.models.esm.tokenization_esm import PreTrainedTokenizer
from transformers import EsmTokenizer, PreTrainedTokenizer


VOCAB_FILES_NAMES = {"vocab_file": "vocab.txt"}


def load_vocab_file(vocab_file):
    with open(vocab_file, "r") as f:
        lines = f.read().splitlines()
        return [l.strip() for l in lines]


class IsoformerTokenizer(PreTrainedTokenizer):
    """
    Constructs Isoformer tokenizer.
    """

    def __init__(
        self,
        **kwargs
    ):
        # Get the model ID from kwargs
        model_id = kwargs.get("name_or_path", None) # This will be "InstaDeepAI/isoformer"

        # Use hf_hub_download to get the local path to each vocabulary file.
        # This function intelligently uses the local cache if the file is already downloaded.
        if model_id:
            try:
                dna_vocab_path = hf_hub_download(repo_id=model_id, filename="dna_vocab_list.txt")
                rna_vocab_path = hf_hub_download(repo_id=model_id, filename="rna_vocab_list.txt")
                protein_vocab_path = hf_hub_download(repo_id=model_id, filename="protein_vocab_list.txt")
            except Exception as e:
                # Fallback in case hf_hub_download fails (e.g., if model_id was a local path not a Hub ID)
                # This fallback might not be perfect for all edge cases, but covers the common local loading.
                print(f"Warning: Failed to resolve model files via hf_hub_download. Attempting local fallback. Error: {e}")
                dna_vocab_path = os.path.join(model_id, "dna_vocab_list.txt")
                rna_vocab_path = os.path.join(model_id, "rna_vocab_list.txt")
                protein_vocab_path = os.path.join(model_id, "protein_vocab_list.txt")
        else:
            # Fallback if model_id is not found (unlikely for AutoTokenizer.from_pretrained)
            print("Warning: Could not determine model_id from kwargs. Falling back to relative paths.")
            dna_vocab_path = "dna_vocab_list.txt"
            rna_vocab_path = "rna_vocab_list.txt"
            protein_vocab_path = "protein_vocab_list.txt"

        dna_hf_tokenizer = EsmTokenizer(dna_vocab_path, model_max_length=196608)
        dna_hf_tokenizer.eos_token = None  # Stops the tokenizer adding an EOS/SEP token at the end
        dna_hf_tokenizer.init_kwargs["eos_token"] = None  # Ensures it doesn't come back when reloading
        dna_hf_tokenizer.bos_token = None  # Stops the tokenizer adding an BOS/SEP token at the end
        dna_hf_tokenizer.init_kwargs["bos_token"] = None  # Ensures it doesn't come back when reloading

        rna_hf_tokenizer = EsmTokenizer(rna_vocab_path, model_max_length=1024)
        rna_hf_tokenizer.eos_token = None  # Stops the tokenizer adding an EOS/SEP token at the end
        rna_hf_tokenizer.init_kwargs["eos_token"] = None  # Ensures it doesn't come back when reloading

        protein_hf_tokenizer = EsmTokenizer(protein_vocab_path, model_max_length=1024)
        # protein_hf_tokenizer.eos_token = None  # Stops the tokenizer adding an EOS/SEP token at the end
        # protein_hf_tokenizer.init_kwargs["eos_token"] = None  # Ensures it doesn't come back when reloading

        self.dna_tokenizer = dna_hf_tokenizer
        self.rna_tokenizer = rna_hf_tokenizer
        self.protein_tokenizer = protein_hf_tokenizer

        self.dna_tokens = open(dna_vocab_path, "r").read() .split("\n")
        self.rna_tokens = open(rna_vocab_path, "r").read() .split("\n")
        self.protein_tokens = open(protein_vocab_path, "r").read() .split("\n")

        super().__init__(**kwargs)

    def __call__(self, dna_input, rna_input, protein_input):
        dna_output = self.dna_tokenizer(dna_input)
        rna_output = self.rna_tokenizer(rna_input, max_length=1024, padding="max_length")
        protein_output = self.protein_tokenizer(protein_input, max_length=1024, padding="max_length")
        return dna_output, rna_output, protein_output

    def _add_tokens(self, *args, **kwargs):
        pass  # Override this with an empty method to stop errors

    def save_vocabulary(self, save_directory, filename_prefix):
        vocab_file_dna = os.path.join(save_directory, (filename_prefix + "-" if filename_prefix else "") + "dna_vocab_list.txt")
        vocab_file_rna = os.path.join(save_directory, (filename_prefix + "-" if filename_prefix else "") + "rna_vocab_list.txt")
        vocab_file_protein = os.path.join(save_directory, (filename_prefix + "-" if filename_prefix else "") + "protein_vocab_list.txt")

        with open(vocab_file_dna, "w") as f:
            f.write("\n".join(self.dna_tokens))
        with open(vocab_file_rna, "w") as f:
            f.write("\n".join(self.rna_tokens))
        with open(vocab_file_protein, "w") as f:
            f.write("\n".join(self.protein_tokens))
        return (vocab_file_dna,vocab_file_rna,vocab_file_protein, )