| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | import pytest |
| 5 | from transformers import AutoTokenizer, PreTrainedTokenizerBase |
| 6 | |
| 7 | from tests.reasoning.utils import run_reasoning_extraction |
| 8 | from vllm.reasoning import ReasoningParser, ReasoningParserManager |
| 9 | |
| 10 | parser_name = "ernie45" |
| 11 | |
| 12 | REASONING_MODEL_NAME = "baidu/ERNIE-4.5-21B-A3B-Thinking" |
| 13 | |
| 14 | |
| 15 | @pytest.fixture(scope="module") |
| 16 | def ernie45_tokenizer() -> PreTrainedTokenizerBase: |
| 17 | return AutoTokenizer.from_pretrained(REASONING_MODEL_NAME) |
| 18 | |
| 19 | |
| 20 | # 带 </think>，非stream |
| 21 | WITH_THINK = { |
| 22 | "output": "abc</think>def", |
| 23 | "reasoning": "abc", |
| 24 | "content": "def", |
| 25 | } |
| 26 | # 带 </think>，stream |
| 27 | WITH_THINK_STREAM = { |
| 28 | "output": "abc</think>def", |
| 29 | "reasoning": "abc", |
| 30 | "content": "def", |
| 31 | } |
| 32 | # without </think>, all is reasoning |
| 33 | WITHOUT_THINK = { |
| 34 | "output": "abc", |
| 35 | "reasoning": "abc", |
| 36 | "content": None, |
| 37 | } |
| 38 | # without </think>, all is reasoning |
| 39 | WITHOUT_THINK_STREAM = { |
| 40 | "output": "abc", |
| 41 | "reasoning": "abc", |
| 42 | "content": None, |
| 43 | } |
| 44 | |
| 45 | COMPLETE_REASONING = { |
| 46 | "output": "abc</think>", |
| 47 | "reasoning": "abc", |
| 48 | "content": None, |
| 49 | } |
| 50 | MULTILINE_REASONING = { |
| 51 | "output": "abc\nABC</think>def\nDEF", |
| 52 | "reasoning": "abc\nABC", |
| 53 | "content": "def\nDEF", |
| 54 | } |
| 55 | |
| 56 | TEST_CASES = [ |
| 57 | pytest.param( |
| 58 | False, |
| 59 | WITH_THINK, |
| 60 | id="with_think", |
| 61 | ), |
| 62 | pytest.param( |
| 63 | True, |
| 64 | WITH_THINK_STREAM, |
| 65 | id="with_think_stream", |
| 66 | ), |
| 67 | pytest.param( |
| 68 | False, |
| 69 | WITHOUT_THINK, |
| 70 | id="without_think", |
| 71 | ), |
| 72 | pytest.param( |
| 73 | True, |
| 74 | WITHOUT_THINK_STREAM, |
| 75 | id="without_think_stream", |
| 76 | ), |
| 77 | pytest.param( |
| 78 | False, |
| 79 | COMPLETE_REASONING, |
| 80 | id="complete_reasoning", |
| 81 | ), |
| 82 | pytest.param( |
| 83 | True, |
| 84 | COMPLETE_REASONING, |
| 85 | id="complete_reasoning_stream", |
| 86 | ), |
| 87 | pytest.param( |
| 88 | False, |
| 89 | MULTILINE_REASONING, |
| 90 | id="multiline_reasoning", |
| 91 | ), |
| 92 | pytest.param( |
| 93 | True, |
| 94 | MULTILINE_REASONING, |
| 95 | id="multiline_reasoning_stream", |
| 96 | ), |
| 97 | ] |
| 98 | |
| 99 | |
| 100 | @pytest.mark.parametrize("streaming, param_dict", TEST_CASES) |
| 101 | def test_reasoning( |
| 102 | streaming: bool, |
| 103 | param_dict: dict, |
| 104 | ernie45_tokenizer, |
| 105 | ): |
| 106 | output = ernie45_tokenizer.tokenize(param_dict["output"]) |
| 107 | output_tokens: list[str] = [] |
| 108 | for token in output: |
| 109 | one_token = ernie45_tokenizer.convert_tokens_to_string([token]) |
| 110 | if one_token: |
| 111 | output_tokens.append(one_token) |
| 112 | |
| 113 | parser: ReasoningParser = ReasoningParserManager.get_reasoning_parser(parser_name)( |
| 114 | ernie45_tokenizer |
| 115 | ) |
| 116 | |
| 117 | reasoning, content = run_reasoning_extraction( |
| 118 | parser, output_tokens, streaming=streaming |
| 119 | ) |
| 120 | |
| 121 | print() |
| 122 | |
| 123 | assert reasoning == param_dict["reasoning"] |
| 124 | assert content == param_dict["content"] |