mirror of
https://github.com/ChatGPTNextWeb/ChatGPT-Next-Web.git
synced 2025-11-18 06:53:41 +08:00
Merge branch 'main' of https://github.com/ChatGPTNextWeb/NextChat into feature-dsmodels-20250228
This commit is contained in:
wangjianhua
287
app/client/platforms/ai302.ts
Normal file
287
app/client/platforms/ai302.ts
Normal file
@@ -0,0 +1,287 @@
|
||||
"use client";
|
||||
|
||||
import {
|
||||
ApiPath,
|
||||
AI302_BASE_URL,
|
||||
DEFAULT_MODELS,
|
||||
AI302,
|
||||
} from "@/app/constant";
|
||||
import {
|
||||
useAccessStore,
|
||||
useAppConfig,
|
||||
useChatStore,
|
||||
ChatMessageTool,
|
||||
usePluginStore,
|
||||
} from "@/app/store";
|
||||
import { preProcessImageContent, streamWithThink } from "@/app/utils/chat";
|
||||
import {
|
||||
ChatOptions,
|
||||
getHeaders,
|
||||
LLMApi,
|
||||
LLMModel,
|
||||
SpeechOptions,
|
||||
} from "../api";
|
||||
import { getClientConfig } from "@/app/config/client";
|
||||
import {
|
||||
getMessageTextContent,
|
||||
getMessageTextContentWithoutThinking,
|
||||
isVisionModel,
|
||||
getTimeoutMSByModel,
|
||||
} from "@/app/utils";
|
||||
import { RequestPayload } from "./openai";
|
||||
|
||||
import { fetch } from "@/app/utils/stream";
|
||||
export interface Ai302ListModelResponse {
|
||||
object: string;
|
||||
data: Array<{
|
||||
id: string;
|
||||
object: string;
|
||||
root: string;
|
||||
}>;
|
||||
}
|
||||
|
||||
export class Ai302Api implements LLMApi {
|
||||
private disableListModels = false;
|
||||
|
||||
path(path: string): string {
|
||||
const accessStore = useAccessStore.getState();
|
||||
|
||||
let baseUrl = "";
|
||||
|
||||
if (accessStore.useCustomConfig) {
|
||||
baseUrl = accessStore.ai302Url;
|
||||
}
|
||||
|
||||
if (baseUrl.length === 0) {
|
||||
const isApp = !!getClientConfig()?.isApp;
|
||||
const apiPath = ApiPath["302.AI"];
|
||||
baseUrl = isApp ? AI302_BASE_URL : apiPath;
|
||||
}
|
||||
|
||||
if (baseUrl.endsWith("/")) {
|
||||
baseUrl = baseUrl.slice(0, baseUrl.length - 1);
|
||||
}
|
||||
if (
|
||||
!baseUrl.startsWith("http") &&
|
||||
!baseUrl.startsWith(ApiPath["302.AI"])
|
||||
) {
|
||||
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 visionModel = isVisionModel(options.config.model);
|
||||
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 = visionModel
|
||||
? await preProcessImageContent(v.content)
|
||||
: getMessageTextContent(v);
|
||||
messages.push({ role: v.role, content });
|
||||
}
|
||||
}
|
||||
|
||||
const modelConfig = {
|
||||
...useAppConfig.getState().modelConfig,
|
||||
...useChatStore.getState().currentSession().mask.modelConfig,
|
||||
...{
|
||||
model: options.config.model,
|
||||
providerName: options.config.providerName,
|
||||
},
|
||||
};
|
||||
|
||||
const requestPayload: RequestPayload = {
|
||||
messages,
|
||||
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(AI302.ChatPath);
|
||||
const chatPayload = {
|
||||
method: "POST",
|
||||
body: JSON.stringify(requestPayload),
|
||||
signal: controller.signal,
|
||||
headers: getHeaders(),
|
||||
};
|
||||
|
||||
// console.log(chatPayload);
|
||||
|
||||
// Use extended timeout for thinking models as they typically require more processing time
|
||||
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[]> {
|
||||
if (this.disableListModels) {
|
||||
return DEFAULT_MODELS.slice();
|
||||
}
|
||||
|
||||
const res = await fetch(this.path(AI302.ListModelPath), {
|
||||
method: "GET",
|
||||
headers: {
|
||||
...getHeaders(),
|
||||
},
|
||||
});
|
||||
|
||||
const resJson = (await res.json()) as Ai302ListModelResponse;
|
||||
const chatModels = resJson.data;
|
||||
console.log("[Models]", chatModels);
|
||||
|
||||
if (!chatModels) {
|
||||
return [];
|
||||
}
|
||||
|
||||
let seq = 1000; //同 Constant.ts 中的排序保持一致
|
||||
return chatModels.map((m) => ({
|
||||
name: m.id,
|
||||
available: true,
|
||||
sorted: seq++,
|
||||
provider: {
|
||||
id: "ai302",
|
||||
providerName: "302.AI",
|
||||
providerType: "ai302",
|
||||
sorted: 15,
|
||||
},
|
||||
}));
|
||||
}
|
||||
}
|
||||
@@ -224,7 +224,7 @@ export class ClaudeApi implements LLMApi {
|
||||
let chunkJson:
|
||||
| undefined
|
||||
| {
|
||||
type: "content_block_delta" | "content_block_stop";
|
||||
type: "content_block_delta" | "content_block_stop" | "message_delta" | "message_stop";
|
||||
content_block?: {
|
||||
type: "tool_use";
|
||||
id: string;
|
||||
@@ -234,11 +234,20 @@ export class ClaudeApi implements LLMApi {
|
||||
type: "text_delta" | "input_json_delta";
|
||||
text?: string;
|
||||
partial_json?: string;
|
||||
stop_reason?: string;
|
||||
};
|
||||
index: number;
|
||||
};
|
||||
chunkJson = JSON.parse(text);
|
||||
|
||||
// Handle refusal stop reason in message_delta
|
||||
if (chunkJson?.delta?.stop_reason === "refusal") {
|
||||
// Return a message to display to the user
|
||||
const refusalMessage = "\n\n[Assistant refused to respond. Please modify your request and try again.]";
|
||||
options.onError?.(new Error("Content policy violation: " + refusalMessage));
|
||||
return refusalMessage;
|
||||
}
|
||||
|
||||
if (chunkJson?.content_block?.type == "tool_use") {
|
||||
index += 1;
|
||||
const id = chunkJson?.content_block.id;
|
||||
|
||||
@@ -56,7 +56,7 @@ export interface OpenAIListModelResponse {
|
||||
|
||||
export interface RequestPayload {
|
||||
messages: {
|
||||
role: "system" | "user" | "assistant";
|
||||
role: "developer" | "system" | "user" | "assistant";
|
||||
content: string | MultimodalContent[];
|
||||
}[];
|
||||
stream?: boolean;
|
||||
@@ -198,7 +198,9 @@ export class ChatGPTApi implements LLMApi {
|
||||
const isDalle3 = _isDalle3(options.config.model);
|
||||
const isO1OrO3 =
|
||||
options.config.model.startsWith("o1") ||
|
||||
options.config.model.startsWith("o3");
|
||||
options.config.model.startsWith("o3") ||
|
||||
options.config.model.startsWith("o4-mini");
|
||||
const isGpt5 = options.config.model.startsWith("gpt-5");
|
||||
if (isDalle3) {
|
||||
const prompt = getMessageTextContent(
|
||||
options.messages.slice(-1)?.pop() as any,
|
||||
@@ -229,7 +231,7 @@ export class ChatGPTApi implements LLMApi {
|
||||
messages,
|
||||
stream: options.config.stream,
|
||||
model: modelConfig.model,
|
||||
temperature: !isO1OrO3 ? modelConfig.temperature : 1,
|
||||
temperature: (!isO1OrO3 && !isGpt5) ? modelConfig.temperature : 1,
|
||||
presence_penalty: !isO1OrO3 ? modelConfig.presence_penalty : 0,
|
||||
frequency_penalty: !isO1OrO3 ? modelConfig.frequency_penalty : 0,
|
||||
top_p: !isO1OrO3 ? modelConfig.top_p : 1,
|
||||
@@ -237,13 +239,28 @@ export class ChatGPTApi implements LLMApi {
|
||||
// Please do not ask me why not send max_tokens, no reason, this param is just shit, I dont want to explain anymore.
|
||||
};
|
||||
|
||||
// O1 使用 max_completion_tokens 控制token数 (https://platform.openai.com/docs/guides/reasoning#controlling-costs)
|
||||
if (isO1OrO3) {
|
||||
if (isGpt5) {
|
||||
// Remove max_tokens if present
|
||||
delete requestPayload.max_tokens;
|
||||
// Add max_completion_tokens (or max_completion_tokens if that's what you meant)
|
||||
requestPayload["max_completion_tokens"] = modelConfig.max_tokens;
|
||||
|
||||
} else if (isO1OrO3) {
|
||||
// by default the o1/o3 models will not attempt to produce output that includes markdown formatting
|
||||
// manually add "Formatting re-enabled" developer message to encourage markdown inclusion in model responses
|
||||
// (https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/reasoning?tabs=python-secure#markdown-output)
|
||||
requestPayload["messages"].unshift({
|
||||
role: "developer",
|
||||
content: "Formatting re-enabled",
|
||||
});
|
||||
|
||||
// o1/o3 uses max_completion_tokens to control the number of tokens (https://platform.openai.com/docs/guides/reasoning#controlling-costs)
|
||||
requestPayload["max_completion_tokens"] = modelConfig.max_tokens;
|
||||
}
|
||||
|
||||
|
||||
// add max_tokens to vision model
|
||||
if (visionModel) {
|
||||
if (visionModel && !isO1OrO3 && ! isGpt5) {
|
||||
requestPayload["max_tokens"] = Math.max(modelConfig.max_tokens, 4000);
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user