调用api接口感觉速度有点慢，是否和流式输出有关系，我看接口记录中流式输出参数是false，这个"stream": false, 参数在哪里可以设置呢？

[zhuifengzl]

2024-04-19 16:22:02,475 - INFO
Client: Address(host='192.168.31.39', port=63902)
Url: http://192.168.31.39:8000/chat/completions
Body: {"max_tokens": 1000, "temperature": 1.2, "top_p": 0.5, "presence_penalty": 0.4, "frequency_penalty": 0.4, "penalty_decay": null, "top_k": null, "global_penalty": null, "messages": [{"role": "user", "content": "喂", "raw": false}], "model": "rwkv", "stream": false, "stop": ["\n\nUser", "\n\nQuestion", "\n\nQ", "\n\nHuman", "\n\nBob", "\n\nAssistant", "\n\nAnswer", "\n\nA", "\n\nBot", "\n\nAlice", "\n\nUser", "\n\nAss"], "user_name": null, "assistant_name": null, "system_name": null, "presystem": true}
Data:
Hello! How can I assist you today?
Finished. RequestsNum: 0

[zhuifengzl]

我用的模型是RWKV-4-World-CHNtuned-3B-v1-20230625-ctx4096.pth，也转换过RWKV-4-World-CHNtuned-3B-v1-20230625-ctx4096-fp16.bin，都测试过，即使修改代码backend-python\routes\completion.py下的设置，重启程序也不生效，是这个模型不支持吗
class ChatCompletionBody(ModelConfigBody):
messages: Union[List[Message], None]
model: Union[str, None] = "rwkv"
stream: bool = False
stop: Union[str, List[str], None] = default_stop
user_name: Union[str, None] = Field(
None, description="Internal user name", min_length=1
)
assistant_name: Union[str, None] = Field(
None, description="Internal assistant name", min_length=1
)
system_name: Union[str, None] = Field(
None, description="Internal system name", min_length=1
)
presystem: bool = Field(
True, description="Whether to insert default system prompt at the beginning"
)

model_config = {
    "json_schema_extra": {
        "example": {
            "messages": [
                {"role": Role.User.value, "content": "hello", "raw": False}
            ],
            "model": "rwkv",
            "stream": False,
            "stop": None,
            "user_name": None,
            "assistant_name": None,
            "system_name": None,
            "presystem": True,
            "max_tokens": 1000,
            "temperature": 1,
            "top_p": 0.3,
            "presence_penalty": 0,
            "frequency_penalty": 1,
        }
    }
}

class CompletionBody(ModelConfigBody):
prompt: Union[str, List[str], None]
model: Union[str, None] = "rwkv"
stream: bool = False
stop: Union[str, List[str], None] = None

model_config = {
    "json_schema_extra": {
        "example": {
            "prompt": "The following is an epic science fiction masterpiece that is immortalized, "
            + "with delicate descriptions and grand depictions of interstellar civilization wars.\nChapter 1.\n",
            "model": "rwkv",
            "stream": False,
            "stop": None,
            "max_tokens": 100,
            "temperature": 1,
            "top_p": 0.3,
            "presence_penalty": 0,
            "frequency_penalty": 1,
        }
    }
}

[zhuifengzl]

好像每次程序都会重置修改的参数，那个流式参数是不能修改吗

[BlinkDL]

模型用 https://huggingface.co/BlinkDL/rwkv-6-world/blob/main/RWKV-x060-World-3B-v2.1-20240417-ctx4096.pth

另外显存多少？显存够就解码参数选 cuda fp16

[zhuifengzl]

12g的显存

[josStorer]

@zhuifengzl 参数是调用api的时候传递的, 可以改, 你不用去改源码, 载入模型的时候, 拉满"载入显存层数"

[zhuifengzl]

好的，我试试，感谢哈

[zhuifengzl]

public struct LocalSendData
{
public string model;
public bool stream;
public bool presystem;
public int max_tokens;
public double temperature;
public double top_p;
public double presence_penalty;
public double frequency_penalty;
public List messages;
}
public struct LocalSendDataMes
{
public string role;
public string content;
public bool raw;
}
public static IEnumerator RequestGPTSegmentation_local(string content, Action<string, bool> callback)
{
string postData = "";
//Debug.Log("postData");

        LocalSendData local = new LocalSendData();
        local.model = "rwkv";
        local.stream = false;
        local.presystem = true;
        local.max_tokens = 10000;
        local.temperature = 1.2;
        local.top_p = 0.5;
        local.presence_penalty = 0.4;
        local.frequency_penalty = 0.4;
        local.messages = new List<LocalSendDataMes>();
        local.messages.Add(new LocalSendDataMes() { role = "user", content = content, raw = false });
        postData = JsonMapper.ToJson(local);

        Debug.Log(postData);

        //Debug.Log(ConfigExcelMgr.instance.excelData.str_localPath);
    
        using (var request = new UnityWebRequest(ConfigExcelMgr.instance.excelData.str_localPath + "/chat/completions", "POST"))
        //using (var request = new UnityWebRequest("http://192.168.0.105:8000/chat/completions", "POST"))
        {
            request.SetRequestHeader("Accept", "application/json, text/plain, */*");
            request.SetRequestHeader("Content-Type", "application/json");                

            request.uploadHandler = new UploadHandlerRaw(Encoding.UTF8.GetBytes(postData));

            request.downloadHandler = new DownloadHandlerBuffer();

            UnityWebRequestAsyncOperation asyncOp = request.SendWebRequest();

            int dataIndex = 0;

            string text = "";

            while (!asyncOp.isDone)
            {
                //Disponse(false);
                yield return wait_internal;
            }
            

            if (request.result == UnityWebRequest.Result.ConnectionError || request.result == UnityWebRequest.Result.ProtocolError)
            {
                Debug.LogError("Error: " + request.error);
            }

            //Debug.Log(request.downloadHandler.text);
            if (request.downloadHandler.text != null)
            {

                string jsondata = request.downloadHandler.text;
                JsonData jd = JsonMapper.ToObject(jsondata);
                //string js_finish_reason = (string)jd["choices"]["finish_reason"];

                string js_text = (string)jd["choices"][0]["message"]["content"];

                callback?.Invoke(js_text, true);
            }

这是写的代码，我不太懂，帮忙看看应该怎么修改，才能提高接口的回答速度呢