| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | import pytest |
| 5 | |
| 6 | from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest |
| 7 | from vllm.entrypoints.openai.completion.protocol import CompletionRequest |
| 8 | from vllm.exceptions import VLLMValidationError |
| 9 | |
| 10 | |
| 11 | @pytest.mark.parametrize("raw_value", [-2, 0.6, 10.5]) |
| 12 | def test_chat_completion_request_rejects_invalid_thinking_token_budget(raw_value): |
| 13 | with pytest.raises(VLLMValidationError, match="thinking_token_budget"): |
| 14 | ChatCompletionRequest.model_validate( |
| 15 | { |
| 16 | "model": "qwen", |
| 17 | "messages": [{"role": "user", "content": "hello"}], |
| 18 | "thinking_token_budget": raw_value, |
| 19 | } |
| 20 | ) |
| 21 | |
| 22 | |
| 23 | def test_chat_completion_request_accepts_valid_thinking_token_budget(): |
| 24 | request = ChatCompletionRequest.model_validate( |
| 25 | { |
| 26 | "model": "qwen", |
| 27 | "messages": [{"role": "user", "content": "hello"}], |
| 28 | "thinking_token_budget": 10, |
| 29 | } |
| 30 | ) |
| 31 | assert request.thinking_token_budget == 10 |
| 32 | |
| 33 | |
| 34 | def test_chat_completion_request_accepts_minus_one_as_unlimited(): |
| 35 | request = ChatCompletionRequest.model_validate( |
| 36 | { |
| 37 | "model": "qwen", |
| 38 | "messages": [{"role": "user", "content": "hello"}], |
| 39 | "thinking_token_budget": -1, |
| 40 | } |
| 41 | ) |
| 42 | assert request.thinking_token_budget is None |
| 43 | |
| 44 | |
| 45 | @pytest.mark.parametrize("raw_value", [0.6, 3.14, -2]) |
| 46 | def test_completion_request_rejects_invalid_thinking_token_budget(raw_value): |
| 47 | with pytest.raises(VLLMValidationError, match="thinking_token_budget"): |
| 48 | CompletionRequest.model_validate( |
| 49 | { |
| 50 | "model": "qwen", |
| 51 | "prompt": "hello", |
| 52 | "thinking_token_budget": raw_value, |
| 53 | } |
| 54 | ) |
| 55 | |
| 56 | |
| 57 | def test_completion_request_accepts_valid_thinking_token_budget(): |
| 58 | request = CompletionRequest.model_validate( |
| 59 | { |
| 60 | "model": "qwen", |
| 61 | "prompt": "hello", |
| 62 | "thinking_token_budget": 5, |
| 63 | } |
| 64 | ) |
| 65 | assert request.thinking_token_budget == 5 |
| 66 | |
| 67 | |
| 68 | def test_completion_request_accepts_minus_one_as_unlimited(): |
| 69 | request = CompletionRequest.model_validate( |
| 70 | { |
| 71 | "model": "qwen", |
| 72 | "prompt": "hello", |
| 73 | "thinking_token_budget": -1, |
| 74 | } |
| 75 | ) |
| 76 | assert request.thinking_token_budget is None |