Chats

约 1621 字大约 5 分钟

2025-04-25

和大模型对话

请求方式: POST

请求地址: /chat/completions

请求参数:

model: 必选 , 模型名, 对应接口 /v1/models 中的 id, 例如 "model": "deepseek-r1:32b-16k"
messages: 必选 , 推理请求消息结构, list类型,0KB<messages内容包含的字符数<4MB,支持中英文。 tokenizer之后的token数量小于或等于maxInputTokenLen, maxSeqLen-1, max_position_embeddings和1MB之间的最小值. 其中,max_position_embeddings从权重文件config.json中获取,其他相关参数从配置文件中获取
1. role: 必选 , 消息角色, 可选值为 "system" 或 "user" 或 "assistant"
提示
deepseek-r1 模型只支持 "user" 类型的消息
1. content: 必选 , 消息内容, 字符串类型
stream: 可选, 是否为流式推理, 默认值为 false, [true: 流式推理, false: 文本推理]
presence_penalty: 可选, 存在惩罚介于-2.0和2.0之间,它影响模型如何根据到目前为止是否出现在文本中来惩罚新token. 正值将通过惩罚已经使用的词,增加模型谈论新主题的可能性, 负值将通过惩罚已经使用的词,减少模型谈论新主题的可能性. 默认值为 0.0, 可选范围[-2.0, 2.0]
frequency_penalty: 可选, 频率惩罚介于-2.0和2.0之间,它影响模型如何根据token的频率来惩罚新token. 正值将通过惩罚常用词,增加模型谈论新主题的可能性, 负值将通过惩罚常用词,减少模型谈论新主题的可能性. 默认值为 0.0, 可选范围[-2.0, 2.0]
repetition_penalty: 可选, 重复惩罚是一种技术,用于减少在文本生成过程中出现重复片段的概率. 它对之前已经生成的文本进行惩罚,使得模型更倾向于选择新的, 不重复的内容. 默认值为 1.0, 可选范围(0.0, 2.0]
temperature: 可选, 温度参数介于0.0和1.0之间,它控制模型生成的文本的创造性. 较低的温度会导致更可预测的文本,而较高的温度会导致更随机的文本. 默认值为 1.0, 可选范围[0.0, 2.0], 0.0表示完全确定性,1.0表示完全随机性, deepseek-r1 模型官方建议设置为 0.6
top_p: 可选, 控制模型生成过程中考虑的词汇范围,使用累计概率选择候选词,直到累计概率超过给定的阈值. 该参数也可以控制生成结果的多样性,它基于累积概率选择候选词,直到累计概率超过给定的阈值为止. 默认值为 1.0, 可选范围(0.0, 1.0]
top_k: 可选, 控制模型生成过程中考虑的词汇范围,只从概率最高的k个候选词中选择. int32类型,取值范围[0, 2147483647],字段未设置时,默认值由后端模型确定
seed: 可选, 用于指定推理过程的随机种子,相同的seed值可以确保推理结果的可重现性,不同的seed值会提升推理结果的随机性. uint_64类型,取值范围[0, 18446744073709551615],不传递该参数,系统会产生一个随机seed值
stop: 可选, 停止推理的token id列表. 输出结果默认不包含停止推理列表中的token id. List[int32]类型,超出int32的元素将会被忽略,默认值null
include_stop_str_in_output: 可选, 决定是否在生成的推理文本中包含停止字符串. bool类型,默认值false。 true: 包含停止字符串. false: 不包含停止字符串. 不传入stop或stop_token_ids时,此字段会被忽略.
skip_special_tokens: 可选, 指定在推理生成的文本中是否跳过特殊tokens. bool类型,默认值true。 true: 跳过特殊token. false: 不跳过特殊token
ignore_eos: 可选, 指定在推理文本生成过程中是否忽略eos_token结束符. bool类型,默认值false。 true: 忽略EOS token. false: 不忽略EOS token
max_length: 可选, 允许推理生成的最大token个数. 实际产生的token数量同时受到配置文件maxIterTimes参数影响,推理token个数小于或等于min(maxIterTimes, max_tokens). int32类型,取值范围[0, 2147483647],默认值为模型设置的maxIterTimes等参数

当所选模型为 qwen3:235b 时, 可通过增加如下参数来控制模型不深度思考:

"chat_template_kwargs": {"enable_thinking": false}

添加这个参数后, 模型将不会进行深度思考, 直接给出回答。如果需要启用深度思考, 则可以将 enable_thinking 设置为 true 或者不设置该参数。

请求和响应示例 - 非流式

curl -X POST "https://uni-api.cstcloud.cn/v1/chat/completions" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer {Token}" \
    -d '{
        "model": "deepseek-r1:32b-16k",
        "messages": [
            {
                "role": "user",
                "content": "Hello!"
            }
        ]
    }'

返回结果

{
    "id": "chatcmpl-702",
    "object": "chat.completion",
    "created": 1739622167,
    "model": "deepseek-r1:32b-16k",
    "system_fingerprint": "fp_ollama",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "&lt;think&gt;\n\n\&lt;/think&gt;\n\nGreetings! I'm DeepSeek-R1, an artificial intelligence assistant created by DeepSeek. I'm at your service and would be delighted to assist you with any inquiries or tasks you may have."
            },
            "finish_reason": "stop"
        }
    ],
    "usage": {
        "prompt_tokens": 6,
        "completion_tokens": 44,
        "total_tokens": 50
    }
}

请求和响应示例 - 流式

curl -X POST "https://uni-api.cstcloud.cn/v1/chat/completions" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer {Token}" \
    -d '{
        "model": "deepseek-r1:32b-16k",
        "messages": [
            {
                "role": "user",
                "content": "Hello!"
            }
        ],
        "stream": true
    }'

返回结果 (server-send-event)

data: {"id": "chatcmpl-949","object": "chat.completion.chunk","created": 1739622469,"model": "deepseek-r1:32b-16k","system_fingerprint": "fp_ollama","choices": [{"index": 0,"delta": {"role": "assistant","content": "&lt;think&gt;"},"finish_reason": null}]}

data: {"id": "chatcmpl-949","object": "chat.completion.chunk","created": 1739622469,"model": "deepseek-r1:32b-16k","system_fingerprint": "fp_ollama","choices": [{"index": 0,"delta": {"role": "assistant","content": "\n\n"},"finish_reason": null}]}

...

data: {"id": "chatcmpl-949","object": "chat.completion.chunk","created": 1739622470,"model": "deepseek-r1:32b-16k","system_fingerprint": "fp_ollama","choices": [],"usage": {"prompt_tokens": 6,"completion_tokens": 44,"total_tokens": 50}}

data: [DONE]

特别提示

当所选模型为 spark-70b-x1 时，流式返回格式如下:

data:{"choices":[{"delta":{"reasoning_content":"\n\n\n","role":"assistant"},"index":1}],"created":1745398469729,"id":"gux5i4h3@spark-gpt-v30@turing-spark@16","model":"xf_IFLYLLM-Spark-X1-ProX@V4.4.2R2.4.0.1_kvxiezai","object":"chat.completion.chunk"}

data:{"choices":[{"delta":{"reasoning_content":"用户发送","role":"assistant"},"index":2}],"created":1745398469775,"id":"gux5i4h3@spark-gpt-v30@turing-spark@16","model":"xf_IFLYLLM-Spark-X1-ProX@V4.4.2R2.4.0.1_kvxiezai","object":"chat.completion.chunk"}

......

data:{"choices":[{"delta":{"reasoning_content":"连贯。\n","role":"assistant"},"index":74}],"created":1745398473762,"id":"gux5i4h3@spark-gpt-v30@turing-spark@16","model":"xf_IFLYLLM-Spark-X1-ProX@V4.4.2R2.4.0.1_kvxiezai","object":"chat.completion.chunk"}

data:{"choices":[{"delta":{"content":"\n\nHello","role":"assistant"},"index":75}],"created":1745398473822,"id":"gux5i4h3@spark-gpt-v30@turing-spark@16","model":"xf_IFLYLLM-Spark-X1-ProX@V4.4.2R2.4.0.1_kvxiezai","object":"chat.completion.chunk"}

data:{"choices":[{"delta":{"content":"! How","role":"assistant"},"index":76}],"created":1745398473878,"id":"gux5i4h3@spark-gpt-v30@turing-spark@16","model":"xf_IFLYLLM-Spark-X1-ProX@V4.4.2R2.4.0.1_kvxiezai","object":"chat.completion.chunk"}

......

data:{"choices":[{"delta":{"content":" here!","role":"assistant"},"index":91}],"created":1745398474747,"id":"gux5i4h3@spark-gpt-v30@turing-spark@16","model":"xf_IFLYLLM-Spark-X1-ProX@V4.4.2R2.4.0.1_kvxiezai","object":"chat.completion.chunk"}

data:{"choices":[{"delta":{"role":"assistant"},"index":92}],"created":1745398474776,"id":"gux5i4h3@spark-gpt-v30@turing-spark@16","model":"xf_IFLYLLM-Spark-X1-ProX@V4.4.2R2.4.0.1_kvxiezai","object":"chat.completion.chunk"}

data:{"choices":[{"delta":{},"finish_reason":"stop","index":92}],"created":1745398474,"id":"gux5i4h3@spark-gpt-v30@turing-spark@16","object":"chat.completion.chunk","usage":{"completion_tokens":35,"prompt_tokens":64,"total_tokens":248}}