审查视图

app/client/platforms/deepseek.ts 7.4 KB
202304001 authored
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253
"use client";
// azure and openai, using same models. so using same LLMApi.
import { ApiPath, DEEPSEEK_BASE_URL, DeepSeek } from "@/app/constant";
import {
  useAccessStore,
  useAppConfig,
  useChatStore,
  ChatMessageTool,
  usePluginStore,
} from "@/app/store";
import { streamWithThink } from "@/app/utils/chat";
import {
  ChatOptions,
  getHeaders,
  LLMApi,
  LLMModel,
  SpeechOptions,
} from "../api";
import { getClientConfig } from "@/app/config/client";
import {
  getMessageTextContent,
  getMessageTextContentWithoutThinking,
  getTimeoutMSByModel,
} from "@/app/utils";
import { RequestPayload } from "./openai";
import { fetch } from "@/app/utils/stream";

export class DeepSeekApi implements LLMApi {
  private disableListModels = true;

  path(path: string): string {
    const accessStore = useAccessStore.getState();

    let baseUrl = "";

    if (accessStore.useCustomConfig) {
      baseUrl = accessStore.deepseekUrl;
    }

    if (baseUrl.length === 0) {
      const isApp = !!getClientConfig()?.isApp;
      const apiPath = ApiPath.DeepSeek;
      baseUrl = isApp ? DEEPSEEK_BASE_URL : apiPath;
    }

    if (baseUrl.endsWith("/")) {
      baseUrl = baseUrl.slice(0, baseUrl.length - 1);
    }
    if (!baseUrl.startsWith("http") && !baseUrl.startsWith(ApiPath.DeepSeek)) {
      baseUrl = "https://" + baseUrl;
    }

    console.log("[Proxy Endpoint] ", baseUrl, path);

    return [baseUrl, path].join("/");
  }

  extractMessage(res: any) {
    return res.choices?.at(0)?.message?.content ?? "";
  }

  speech(options: SpeechOptions): Promise<ArrayBuffer> {
    throw new Error("Method not implemented.");
  }

  async chat(options: ChatOptions) {
    const messages: ChatOptions["messages"] = [];
    for (const v of options.messages) {
      if (v.role === "assistant") {
        const content = getMessageTextContentWithoutThinking(v);
        messages.push({ role: v.role, content });
      } else {
        const content = getMessageTextContent(v);
        messages.push({ role: v.role, content });
      }
    }

    // 检测并修复消息顺序,确保除system外的第一个消息是user
    const filteredMessages: ChatOptions["messages"] = [];
    let hasFoundFirstUser = false;

    for (const msg of messages) {
      if (msg.role === "system") {
        // Keep all system messages
        filteredMessages.push(msg);
      } else if (msg.role === "user") {
        // User message directly added
        filteredMessages.push(msg);
        hasFoundFirstUser = true;
      } else if (hasFoundFirstUser) {
        // After finding the first user message, all subsequent non-system messages are retained.
        filteredMessages.push(msg);
      }
      // If hasFoundFirstUser is false and it is not a system message, it will be skipped.
    }

    const modelConfig = {
      ...useAppConfig.getState().modelConfig,
      ...useChatStore.getState().currentSession().mask.modelConfig,
      ...{
        model: options.config.model,
        providerName: options.config.providerName,
      },
    };

    const requestPayload: RequestPayload = {
      messages: filteredMessages,
      stream: options.config.stream,
      model: modelConfig.model,
      temperature: modelConfig.temperature,
      presence_penalty: modelConfig.presence_penalty,
      frequency_penalty: modelConfig.frequency_penalty,
      top_p: modelConfig.top_p,
      // max_tokens: Math.max(modelConfig.max_tokens, 1024),
      // Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
    };

    console.log("[Request] openai payload: ", requestPayload);

    const shouldStream = !!options.config.stream;
    const controller = new AbortController();
    options.onController?.(controller);

    try {
      const chatPath = this.path(DeepSeek.ChatPath);
      const chatPayload = {
        method: "POST",
        body: JSON.stringify(requestPayload),
        signal: controller.signal,
        headers: getHeaders(),
      };

      // make a fetch request
      const requestTimeoutId = setTimeout(
        () => controller.abort(),
        getTimeoutMSByModel(options.config.model),
      );

      if (shouldStream) {
        const [tools, funcs] = usePluginStore
          .getState()
          .getAsTools(
            useChatStore.getState().currentSession().mask?.plugin || [],
          );
        return streamWithThink(
          chatPath,
          requestPayload,
          getHeaders(),
          tools as any,
          funcs,
          controller,
          // parseSSE
          (text: string, runTools: ChatMessageTool[]) => {
            // console.log("parseSSE", text, runTools);
            const json = JSON.parse(text);
            const choices = json.choices as Array<{
              delta: {
                content: string | null;
                tool_calls: ChatMessageTool[];
                reasoning_content: string | null;
              };
            }>;
            const tool_calls = choices[0]?.delta?.tool_calls;
            if (tool_calls?.length > 0) {
              const index = tool_calls[0]?.index;
              const id = tool_calls[0]?.id;
              const args = tool_calls[0]?.function?.arguments;
              if (id) {
                runTools.push({
                  id,
                  type: tool_calls[0]?.type,
                  function: {
                    name: tool_calls[0]?.function?.name as string,
                    arguments: args,
                  },
                });
              } else {
                // @ts-ignore
                runTools[index]["function"]["arguments"] += args;
              }
            }
            const reasoning = choices[0]?.delta?.reasoning_content;
            const content = choices[0]?.delta?.content;

            // Skip if both content and reasoning_content are empty or null
            if (
              (!reasoning || reasoning.length === 0) &&
              (!content || content.length === 0)
            ) {
              return {
                isThinking: false,
                content: "",
              };
            }

            if (reasoning && reasoning.length > 0) {
              return {
                isThinking: true,
                content: reasoning,
              };
            } else if (content && content.length > 0) {
              return {
                isThinking: false,
                content: content,
              };
            }

            return {
              isThinking: false,
              content: "",
            };
          },
          // processToolMessage, include tool_calls message and tool call results
          (
            requestPayload: RequestPayload,
            toolCallMessage: any,
            toolCallResult: any[],
          ) => {
            // @ts-ignore
            requestPayload?.messages?.splice(
              // @ts-ignore
              requestPayload?.messages?.length,
              0,
              toolCallMessage,
              ...toolCallResult,
            );
          },
          options,
        );
      } else {
        const res = await fetch(chatPath, chatPayload);
        clearTimeout(requestTimeoutId);

        const resJson = await res.json();
        const message = this.extractMessage(resJson);
        options.onFinish(message, res);
      }
    } catch (e) {
      console.log("[Request] failed to make a chat request", e);
      options.onError?.(e as Error);
    }
  }
  async usage() {
    return {
      used: 0,
      total: 0,
    };
  }

  async models(): Promise<LLMModel[]> {
    return [];
  }
}