| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | from unittest.mock import MagicMock |
| 5 | |
| 6 | import pytest |
| 7 | |
| 8 | from vllm.entrypoints.openai.chat_completion.protocol import ( |
| 9 | ChatCompletionRequest, |
| 10 | ) |
| 11 | |
| 12 | |
| 13 | @pytest.fixture() |
| 14 | def should_do_global_cleanup_after_test() -> bool: |
| 15 | return False |
| 16 | |
| 17 | |
| 18 | def make_mock_tokenizer( |
| 19 | vocab: dict[str, int], |
| 20 | special_tokens: list[str] | None = None, |
| 21 | ) -> MagicMock: |
| 22 | """Create a mock tokenizer with the given vocabulary. |
| 23 | |
| 24 | Args: |
| 25 | vocab: Mapping of token text to token ID. |
| 26 | special_tokens: Which tokens to mark as special. When ``None`` |
| 27 | (the default), every key in *vocab* is treated as special — |
| 28 | convenient when the vocab only contains delimiter tokens. |
| 29 | |
| 30 | The returned mock supports get_vocab(), encode(), and decode(). |
| 31 | decode() maps known token IDs back to their text and falls back to |
| 32 | chr(id) for ASCII IDs or ``<id>`` for others. |
| 33 | |
| 34 | """ |
| 35 | id_to_text = {v: k for k, v in vocab.items()} |
| 36 | tokenizer = MagicMock() |
| 37 | tokenizer.encode.return_value = [1, 2, 3] |
| 38 | tokenizer.get_vocab.return_value = dict(vocab) |
| 39 | tokenizer.decode.side_effect = lambda ids: "".join( |
| 40 | id_to_text.get(i, chr(i) if i < 128 else f"<{i}>") for i in ids |
| 41 | ) |
| 42 | st = special_tokens if special_tokens is not None else list(vocab.keys()) |
| 43 | tokenizer.all_special_tokens = st |
| 44 | tokenizer.all_special_ids = [vocab[t] for t in st if t in vocab] |
| 45 | return tokenizer |
| 46 | |
| 47 | |
| 48 | @pytest.fixture |
| 49 | def mock_request(): |
| 50 | req = MagicMock(spec=ChatCompletionRequest) |
| 51 | req.tools = [] |
| 52 | req.tool_choice = "auto" |
| 53 | req.include_reasoning = True |
| 54 | return req |