Function test_unstream_with_logprobs

tests/ce/server/test_logprobs.py:6–38

测试非流式响应开启 logprobs 后，返回的 token 概率信息是否正确。

()

def test_unstream_with_logprobs():
    """
    测试非流式响应开启 logprobs 后，返回的 token 概率信息是否正确。
    """
    data = {
        "stream": False,
        "messages": [
            {"role": "system", "content": "You are a helpful assistant."},
            {"role": "user", "content": "牛顿的三大运动定律是什么？"},
        ],
        "max_tokens": 3,
    }

    # 构建请求并发送
    payload = build_request_payload(TEMPLATE, data)
    response = send_request(URL, payload)
    print(json.dumps(response.json(), indent=2, ensure_ascii=False))
    resp_json = response.json()

    # 校验返回内容与概率信息
    assert resp_json["choices"][0]["message"]["content"] == "牛顿的"
    assert resp_json["choices"][0]["logprobs"]["content"][0]["token"] == "牛顿"
    assert resp_json["choices"][0]["logprobs"]["content"][0]["logprob"] == -0.031025361269712448
    assert resp_json["choices"][0]["logprobs"]["content"][0]["top_logprobs"][0] == {
        "token": "牛顿",
        "logprob": -0.031025361269712448,
        "bytes": [231, 137, 155, 233, 161, 191],
        "top_logprobs": None,
    }

    assert resp_json["usage"]["prompt_tokens"] == 22
    assert resp_json["usage"]["completion_tokens"] == 3
    assert resp_json["usage"]["total_tokens"] == 25


def test_unstream_without_logprobs():