| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | import openai # use the official client for correctness check |
| 5 | import openai.types.responses as openai_responses_types |
| 6 | import pytest |
| 7 | |
| 8 | |
| 9 | @pytest.mark.asyncio |
| 10 | async def test_simple_input(client: openai.AsyncOpenAI): |
| 11 | response = await client.responses.create(input="What is 13 * 24?") |
| 12 | print(response) |
| 13 | |
| 14 | outputs = response.output |
| 15 | # Whether the output contains the answer. |
| 16 | assert outputs[-1].type == "message" |
| 17 | assert "312" in outputs[-1].content[0].text |
| 18 | |
| 19 | # Whether the output contains the reasoning. |
| 20 | assert outputs[0].type == "reasoning" |
| 21 | assert outputs[0].content[0].text != "" |
| 22 | |
| 23 | |
| 24 | @pytest.mark.asyncio |
| 25 | async def test_instructions(client: openai.AsyncOpenAI): |
| 26 | response = await client.responses.create( |
| 27 | instructions="Finish the answer with QED.", |
| 28 | input="What is 13 * 24?", |
| 29 | ) |
| 30 | print(response) |
| 31 | |
| 32 | output_text = response.output[-1].content[0].text |
| 33 | assert "312" in output_text |
| 34 | assert "QED" in output_text |
| 35 | |
| 36 | |
| 37 | @pytest.mark.asyncio |
| 38 | async def test_chat(client: openai.AsyncOpenAI): |
| 39 | response = await client.responses.create( |
| 40 | input=[ |
| 41 | {"role": "system", "content": "Finish the answer with QED."}, |
| 42 | {"role": "user", "content": "What is 5 * 3?"}, |
| 43 | {"role": "assistant", "content": "15. QED."}, |
| 44 | {"role": "user", "content": "Multiply the result by 2."}, |
| 45 | ], |
| 46 | ) |
| 47 | print(response) |
| 48 | |
| 49 | output_text = response.output[-1].content[0].text |
| 50 | assert "30" in output_text |
| 51 | assert "QED" in output_text |
| 52 | |
| 53 | |
| 54 | @pytest.mark.asyncio |
| 55 | async def test_chat_with_input_type(client: openai.AsyncOpenAI): |
| 56 | response = await client.responses.create( |
| 57 | input=[ |
| 58 | { |
| 59 | "role": "user", |
| 60 | "content": [{"type": "input_text", "text": "Hello!"}], |
| 61 | }, |
| 62 | ], |
| 63 | ) |
| 64 | print(response) |
| 65 | assert response.status == "completed" |
| 66 | |
| 67 | |
| 68 | @pytest.mark.asyncio |
| 69 | async def test_logprobs(client: openai.AsyncOpenAI): |
| 70 | response = await client.responses.create( |
| 71 | include=["message.output_text.logprobs"], |
| 72 | input="What is 13 * 24?", |
| 73 | top_logprobs=5, |
| 74 | ) |
| 75 | print(response) |
| 76 | outputs = response.output |
| 77 | assert outputs[-1].content[-1].logprobs |
| 78 | assert len(outputs[-1].content[-1].logprobs[0].top_logprobs) == 5 |
| 79 | |
| 80 | |
| 81 | @pytest.mark.asyncio |
| 82 | async def test_streaming(client: openai.AsyncOpenAI): |
| 83 | stream = await client.responses.create( |
| 84 | input="What is 13 * 24?", |
| 85 | stream=True, |
| 86 | ) |
| 87 | events = [event async for event in stream] |
| 88 | assert isinstance(events[0], openai_responses_types.ResponseCreatedEvent) |
| 89 | assert any( |
| 90 | isinstance(event, openai_responses_types.ResponseTextDeltaEvent) |
| 91 | for event in events |
| 92 | ) |
| 93 | assert isinstance(events[-1], openai_responses_types.ResponseCompletedEvent) |