test_tokenizer_group.py

# SPDX-License-Identifier: Apache-2.0

import os
import pytest
from transformers import AutoTokenizer, PreTrainedTokenizerBase
from ..utils import models_path_prefix

from vllm.transformers_utils.tokenizer_group import TokenizerGroup


# export HF_ENDPOINT=https://hf-mirror.com
@pytest.mark.asyncio
async def test_tokenizer_group():
    # reference_tokenizer = AutoTokenizer.from_pretrained(os.path.join(models_path_prefix, "gpt2"))
    reference_tokenizer = AutoTokenizer.from_pretrained("gpt2")
    tokenizer_group = TokenizerGroup(
        # tokenizer_id=os.path.join(models_path_prefix, "gpt2"),
        enable_lora=False,
        max_num_seqs=1,
        max_input_length=None,
    )
    assert reference_tokenizer.encode("prompt") == tokenizer_group.encode(
        prompt="prompt", lora_request=None)
    assert reference_tokenizer.encode(
        "prompt") == await tokenizer_group.encode_async(prompt="prompt",
                                                        lora_request=None)
    assert isinstance(tokenizer_group.get_lora_tokenizer(None),
                      PreTrainedTokenizerBase)
    assert tokenizer_group.get_lora_tokenizer(
        None) == await tokenizer_group.get_lora_tokenizer_async(None)