622 lines
19 KiB
TypeScript
622 lines
19 KiB
TypeScript
"use client";
|
||
// azure and openai, using same models. so using same LLMApi.
|
||
import {
|
||
ApiPath,
|
||
DEFAULT_API_HOST,
|
||
DEFAULT_MODELS,
|
||
OpenaiPath,
|
||
Azure,
|
||
REQUEST_TIMEOUT_MS,
|
||
ServiceProvider,
|
||
} from "@/app/constant";
|
||
import {
|
||
ChatMessageTool,
|
||
useAccessStore,
|
||
useAppConfig,
|
||
useChatStore,
|
||
} from "@/app/store";
|
||
import { collectModelsWithDefaultModel } from "@/app/utils/model";
|
||
import {
|
||
preProcessImageContent,
|
||
uploadImage,
|
||
base64Image2Blob,
|
||
} from "@/app/utils/chat";
|
||
import { cloudflareAIGatewayUrl } from "@/app/utils/cloudflare";
|
||
import { DalleSize, DalleQuality, DalleStyle } from "@/app/typing";
|
||
|
||
import {
|
||
ChatOptions,
|
||
getHeaders,
|
||
LLMApi,
|
||
LLMModel,
|
||
LLMUsage,
|
||
MultimodalContent,
|
||
} from "../api";
|
||
import Locale from "../../locales";
|
||
import {
|
||
EventStreamContentType,
|
||
fetchEventSource,
|
||
} from "@fortaine/fetch-event-source";
|
||
import { prettyObject } from "@/app/utils/format";
|
||
import { getClientConfig } from "@/app/config/client";
|
||
import {
|
||
getMessageTextContent,
|
||
getMessageImages,
|
||
isVisionModel,
|
||
isDalle3 as _isDalle3,
|
||
} from "@/app/utils";
|
||
|
||
export interface OpenAIListModelResponse {
|
||
object: string;
|
||
data: Array<{
|
||
id: string;
|
||
object: string;
|
||
root: string;
|
||
}>;
|
||
}
|
||
|
||
export interface RequestPayload {
|
||
messages: {
|
||
role: "system" | "user" | "assistant";
|
||
content: string | MultimodalContent[];
|
||
}[];
|
||
stream?: boolean;
|
||
model: string;
|
||
temperature: number;
|
||
presence_penalty: number;
|
||
frequency_penalty: number;
|
||
top_p: number;
|
||
max_tokens?: number;
|
||
}
|
||
|
||
export interface DalleRequestPayload {
|
||
model: string;
|
||
prompt: string;
|
||
response_format: "url" | "b64_json";
|
||
n: number;
|
||
size: DalleSize;
|
||
quality: DalleQuality;
|
||
style: DalleStyle;
|
||
}
|
||
|
||
export class ChatGPTApi implements LLMApi {
|
||
private disableListModels = true;
|
||
|
||
path(path: string): string {
|
||
const accessStore = useAccessStore.getState();
|
||
|
||
let baseUrl = "";
|
||
|
||
const isAzure = path.includes("deployments");
|
||
if (accessStore.useCustomConfig) {
|
||
if (isAzure && !accessStore.isValidAzure()) {
|
||
throw Error(
|
||
"incomplete azure config, please check it in your settings page",
|
||
);
|
||
}
|
||
|
||
baseUrl = isAzure ? accessStore.azureUrl : accessStore.openaiUrl;
|
||
}
|
||
|
||
if (baseUrl.length === 0) {
|
||
const isApp = !!getClientConfig()?.isApp;
|
||
const apiPath = isAzure ? ApiPath.Azure : ApiPath.OpenAI;
|
||
baseUrl = isApp ? DEFAULT_API_HOST + "/proxy" + apiPath : apiPath;
|
||
}
|
||
|
||
if (baseUrl.endsWith("/")) {
|
||
baseUrl = baseUrl.slice(0, baseUrl.length - 1);
|
||
}
|
||
if (
|
||
!baseUrl.startsWith("http") &&
|
||
!isAzure &&
|
||
!baseUrl.startsWith(ApiPath.OpenAI)
|
||
) {
|
||
baseUrl = "https://" + baseUrl;
|
||
}
|
||
|
||
console.log("[Proxy Endpoint] ", baseUrl, path);
|
||
|
||
// try rebuild url, when using cloudflare ai gateway in client
|
||
return cloudflareAIGatewayUrl([baseUrl, path].join("/"));
|
||
}
|
||
|
||
async extractMessage(res: any) {
|
||
if (res.error) {
|
||
return "```\n" + JSON.stringify(res, null, 4) + "\n```";
|
||
}
|
||
// dalle3 model return url, using url create image message
|
||
if (res.data) {
|
||
let url = res.data?.at(0)?.url ?? "";
|
||
const b64_json = res.data?.at(0)?.b64_json ?? "";
|
||
if (!url && b64_json) {
|
||
// uploadImage
|
||
url = await uploadImage(base64Image2Blob(b64_json, "image/png"));
|
||
}
|
||
return [
|
||
{
|
||
type: "image_url",
|
||
image_url: {
|
||
url,
|
||
},
|
||
},
|
||
];
|
||
}
|
||
return res.choices?.at(0)?.message?.content ?? res;
|
||
}
|
||
|
||
async chat(options: ChatOptions) {
|
||
const modelConfig = {
|
||
...useAppConfig.getState().modelConfig,
|
||
...useChatStore.getState().currentSession().mask.modelConfig,
|
||
...{
|
||
model: options.config.model,
|
||
providerName: options.config.providerName,
|
||
},
|
||
};
|
||
|
||
let requestPayload: RequestPayload | DalleRequestPayload;
|
||
|
||
const isDalle3 = _isDalle3(options.config.model);
|
||
if (isDalle3) {
|
||
const prompt = getMessageTextContent(
|
||
options.messages.slice(-1)?.pop() as any,
|
||
);
|
||
requestPayload = {
|
||
model: options.config.model,
|
||
prompt,
|
||
// URLs are only valid for 60 minutes after the image has been generated.
|
||
response_format: "b64_json", // using b64_json, and save image in CacheStorage
|
||
n: 1,
|
||
size: options.config?.size ?? "1024x1024",
|
||
quality: options.config?.quality ?? "standard",
|
||
style: options.config?.style ?? "vivid",
|
||
};
|
||
} else {
|
||
const visionModel = isVisionModel(options.config.model);
|
||
const messages: ChatOptions["messages"] = [];
|
||
for (const v of options.messages) {
|
||
const content = visionModel
|
||
? await preProcessImageContent(v.content)
|
||
: getMessageTextContent(v);
|
||
messages.push({ role: v.role, content });
|
||
}
|
||
|
||
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.
|
||
};
|
||
|
||
// add max_tokens to vision model
|
||
if (visionModel && modelConfig.model.includes("preview")) {
|
||
requestPayload["max_tokens"] = Math.max(modelConfig.max_tokens, 4000);
|
||
}
|
||
}
|
||
|
||
console.log("[Request] openai payload: ", requestPayload);
|
||
|
||
const shouldStream = !isDalle3 && !!options.config.stream;
|
||
const controller = new AbortController();
|
||
options.onController?.(controller);
|
||
|
||
try {
|
||
let chatPath = "";
|
||
if (modelConfig.providerName === ServiceProvider.Azure) {
|
||
// find model, and get displayName as deployName
|
||
const { models: configModels, customModels: configCustomModels } =
|
||
useAppConfig.getState();
|
||
const {
|
||
defaultModel,
|
||
customModels: accessCustomModels,
|
||
useCustomConfig,
|
||
} = useAccessStore.getState();
|
||
const models = collectModelsWithDefaultModel(
|
||
configModels,
|
||
[configCustomModels, accessCustomModels].join(","),
|
||
defaultModel,
|
||
);
|
||
const model = models.find(
|
||
(model) =>
|
||
model.name === modelConfig.model &&
|
||
model?.provider?.providerName === ServiceProvider.Azure,
|
||
);
|
||
chatPath = this.path(
|
||
(isDalle3 ? Azure.ImagePath : Azure.ChatPath)(
|
||
(model?.displayName ?? model?.name) as string,
|
||
useCustomConfig ? useAccessStore.getState().azureApiVersion : "",
|
||
),
|
||
);
|
||
} else {
|
||
chatPath = this.path(
|
||
isDalle3 ? OpenaiPath.ImagePath : OpenaiPath.ChatPath,
|
||
);
|
||
}
|
||
const chatPayload = {
|
||
method: "POST",
|
||
body: JSON.stringify(requestPayload),
|
||
signal: controller.signal,
|
||
headers: getHeaders(),
|
||
};
|
||
|
||
// make a fetch request
|
||
const requestTimeoutId = setTimeout(
|
||
() => controller.abort(),
|
||
isDalle3 ? REQUEST_TIMEOUT_MS * 2 : REQUEST_TIMEOUT_MS, // dalle3 using b64_json is slow.
|
||
);
|
||
|
||
if (shouldStream) {
|
||
let responseText = "";
|
||
let remainText = "";
|
||
let finished = false;
|
||
let running = false;
|
||
let runTools: ChatMessageTool[] = [];
|
||
|
||
// animate response to make it looks smooth
|
||
function animateResponseText() {
|
||
if (finished || controller.signal.aborted) {
|
||
responseText += remainText;
|
||
console.log("[Response Animation] finished");
|
||
if (responseText?.length === 0) {
|
||
options.onError?.(new Error("empty response from server"));
|
||
}
|
||
return;
|
||
}
|
||
|
||
if (remainText.length > 0) {
|
||
const fetchCount = Math.max(1, Math.round(remainText.length / 60));
|
||
const fetchText = remainText.slice(0, fetchCount);
|
||
responseText += fetchText;
|
||
remainText = remainText.slice(fetchCount);
|
||
options.onUpdate?.(responseText, fetchText);
|
||
}
|
||
|
||
requestAnimationFrame(animateResponseText);
|
||
}
|
||
|
||
// start animaion
|
||
animateResponseText();
|
||
|
||
// TODO 后面这里是从选择的plugins中获取function列表
|
||
const funcs = {
|
||
get_current_weather: (args: any) => {
|
||
console.log("call get_current_weather", args);
|
||
return "30";
|
||
},
|
||
};
|
||
const finish = () => {
|
||
if (!finished) {
|
||
console.log("try run tools", runTools.length, finished, running);
|
||
if (!running && runTools.length > 0) {
|
||
const toolCallMessage = {
|
||
role: "assistant",
|
||
tool_calls: [...runTools],
|
||
};
|
||
running = true;
|
||
runTools.splice(0, runTools.length); // empty runTools
|
||
return Promise.all(
|
||
toolCallMessage.tool_calls.map((tool) => {
|
||
options?.onBeforeTool?.(tool);
|
||
return Promise.resolve(
|
||
// @ts-ignore
|
||
funcs[tool.function.name](
|
||
// @ts-ignore
|
||
JSON.parse(tool.function.arguments),
|
||
),
|
||
)
|
||
.then((content) => {
|
||
options?.onAfterTool?.({
|
||
...tool,
|
||
content,
|
||
isError: false,
|
||
});
|
||
return content;
|
||
})
|
||
.catch((e) => {
|
||
options?.onAfterTool?.({ ...tool, isError: true });
|
||
return e.toString();
|
||
})
|
||
.then((content) => ({
|
||
role: "tool",
|
||
content,
|
||
tool_call_id: tool.id,
|
||
}));
|
||
}),
|
||
).then((toolCallResult) => {
|
||
console.log("end runTools", toolCallMessage, toolCallResult);
|
||
// @ts-ignore
|
||
requestPayload?.messages?.splice(
|
||
// @ts-ignore
|
||
requestPayload?.messages?.length,
|
||
0,
|
||
toolCallMessage,
|
||
...toolCallResult,
|
||
);
|
||
setTimeout(() => {
|
||
// call again
|
||
console.log("start again");
|
||
running = false;
|
||
chatApi(chatPath, requestPayload as RequestPayload); // call fetchEventSource
|
||
}, 0);
|
||
});
|
||
console.log("try run tools", runTools.length, finished);
|
||
return;
|
||
}
|
||
if (running) {
|
||
return;
|
||
}
|
||
finished = true;
|
||
options.onFinish(responseText + remainText);
|
||
}
|
||
};
|
||
|
||
controller.signal.onabort = finish;
|
||
|
||
function chatApi(chatPath: string, requestPayload: RequestPayload) {
|
||
const chatPayload = {
|
||
method: "POST",
|
||
body: JSON.stringify({
|
||
...requestPayload,
|
||
// TODO 这里暂时写死的，后面从store.tools中按照当前session中选择的获取
|
||
tools: [
|
||
{
|
||
type: "function",
|
||
function: {
|
||
name: "get_current_weather",
|
||
description: "Get the current weather",
|
||
parameters: {
|
||
type: "object",
|
||
properties: {
|
||
location: {
|
||
type: "string",
|
||
description:
|
||
"The city and country, eg. San Francisco, USA",
|
||
},
|
||
format: {
|
||
type: "string",
|
||
enum: ["celsius", "fahrenheit"],
|
||
},
|
||
},
|
||
required: ["location", "format"],
|
||
},
|
||
},
|
||
},
|
||
],
|
||
}),
|
||
signal: controller.signal,
|
||
headers: getHeaders(),
|
||
};
|
||
console.log("chatApi", chatPath, requestPayload, chatPayload);
|
||
fetchEventSource(chatPath, {
|
||
...chatPayload,
|
||
async onopen(res) {
|
||
clearTimeout(requestTimeoutId);
|
||
const contentType = res.headers.get("content-type");
|
||
console.log(
|
||
"[OpenAI] request response content type: ",
|
||
contentType,
|
||
);
|
||
|
||
if (contentType?.startsWith("text/plain")) {
|
||
responseText = await res.clone().text();
|
||
return finish();
|
||
}
|
||
|
||
if (
|
||
!res.ok ||
|
||
!res.headers
|
||
.get("content-type")
|
||
?.startsWith(EventStreamContentType) ||
|
||
res.status !== 200
|
||
) {
|
||
const responseTexts = [responseText];
|
||
let extraInfo = await res.clone().text();
|
||
try {
|
||
const resJson = await res.clone().json();
|
||
extraInfo = prettyObject(resJson);
|
||
} catch {}
|
||
|
||
if (res.status === 401) {
|
||
responseTexts.push(Locale.Error.Unauthorized);
|
||
}
|
||
|
||
if (extraInfo) {
|
||
responseTexts.push(extraInfo);
|
||
}
|
||
|
||
responseText = responseTexts.join("\n\n");
|
||
|
||
return finish();
|
||
}
|
||
},
|
||
onmessage(msg) {
|
||
if (msg.data === "[DONE]" || finished) {
|
||
return finish();
|
||
}
|
||
const text = msg.data;
|
||
try {
|
||
const json = JSON.parse(text);
|
||
const choices = json.choices as Array<{
|
||
delta: {
|
||
content: string;
|
||
tool_calls: ChatMessageTool[];
|
||
};
|
||
}>;
|
||
console.log("choices", choices);
|
||
const delta = choices[0]?.delta?.content;
|
||
const tool_calls = choices[0]?.delta?.tool_calls;
|
||
const textmoderation = json?.prompt_filter_results;
|
||
|
||
if (delta) {
|
||
remainText += delta;
|
||
}
|
||
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;
|
||
}
|
||
}
|
||
|
||
console.log("runTools", runTools);
|
||
|
||
if (
|
||
textmoderation &&
|
||
textmoderation.length > 0 &&
|
||
ServiceProvider.Azure
|
||
) {
|
||
const contentFilterResults =
|
||
textmoderation[0]?.content_filter_results;
|
||
console.log(
|
||
`[${ServiceProvider.Azure}] [Text Moderation] flagged categories result:`,
|
||
contentFilterResults,
|
||
);
|
||
}
|
||
} catch (e) {
|
||
console.error("[Request] parse error", text, msg);
|
||
}
|
||
},
|
||
onclose() {
|
||
finish();
|
||
},
|
||
onerror(e) {
|
||
options.onError?.(e);
|
||
throw e;
|
||
},
|
||
openWhenHidden: true,
|
||
});
|
||
}
|
||
chatApi(chatPath, requestPayload as RequestPayload); // call fetchEventSource
|
||
} else {
|
||
const res = await fetch(chatPath, chatPayload);
|
||
clearTimeout(requestTimeoutId);
|
||
|
||
const resJson = await res.json();
|
||
const message = await this.extractMessage(resJson);
|
||
options.onFinish(message);
|
||
}
|
||
} catch (e) {
|
||
console.log("[Request] failed to make a chat request", e);
|
||
options.onError?.(e as Error);
|
||
}
|
||
}
|
||
async usage() {
|
||
const formatDate = (d: Date) =>
|
||
`${d.getFullYear()}-${(d.getMonth() + 1).toString().padStart(2, "0")}-${d
|
||
.getDate()
|
||
.toString()
|
||
.padStart(2, "0")}`;
|
||
const ONE_DAY = 1 * 24 * 60 * 60 * 1000;
|
||
const now = new Date();
|
||
const startOfMonth = new Date(now.getFullYear(), now.getMonth(), 1);
|
||
const startDate = formatDate(startOfMonth);
|
||
const endDate = formatDate(new Date(Date.now() + ONE_DAY));
|
||
|
||
const [used, subs] = await Promise.all([
|
||
fetch(
|
||
this.path(
|
||
`${OpenaiPath.UsagePath}?start_date=${startDate}&end_date=${endDate}`,
|
||
),
|
||
{
|
||
method: "GET",
|
||
headers: getHeaders(),
|
||
},
|
||
),
|
||
fetch(this.path(OpenaiPath.SubsPath), {
|
||
method: "GET",
|
||
headers: getHeaders(),
|
||
}),
|
||
]);
|
||
|
||
if (used.status === 401) {
|
||
throw new Error(Locale.Error.Unauthorized);
|
||
}
|
||
|
||
if (!used.ok || !subs.ok) {
|
||
throw new Error("Failed to query usage from openai");
|
||
}
|
||
|
||
const response = (await used.json()) as {
|
||
total_usage?: number;
|
||
error?: {
|
||
type: string;
|
||
message: string;
|
||
};
|
||
};
|
||
|
||
const total = (await subs.json()) as {
|
||
hard_limit_usd?: number;
|
||
};
|
||
|
||
if (response.error && response.error.type) {
|
||
throw Error(response.error.message);
|
||
}
|
||
|
||
if (response.total_usage) {
|
||
response.total_usage = Math.round(response.total_usage) / 100;
|
||
}
|
||
|
||
if (total.hard_limit_usd) {
|
||
total.hard_limit_usd = Math.round(total.hard_limit_usd * 100) / 100;
|
||
}
|
||
|
||
return {
|
||
used: response.total_usage,
|
||
total: total.hard_limit_usd,
|
||
} as LLMUsage;
|
||
}
|
||
|
||
async models(): Promise<LLMModel[]> {
|
||
if (this.disableListModels) {
|
||
return DEFAULT_MODELS.slice();
|
||
}
|
||
|
||
const res = await fetch(this.path(OpenaiPath.ListModelPath), {
|
||
method: "GET",
|
||
headers: {
|
||
...getHeaders(),
|
||
},
|
||
});
|
||
|
||
const resJson = (await res.json()) as OpenAIListModelResponse;
|
||
const chatModels = resJson.data?.filter((m) => m.id.startsWith("gpt-"));
|
||
console.log("[Models]", chatModels);
|
||
|
||
if (!chatModels) {
|
||
return [];
|
||
}
|
||
|
||
//由于目前 OpenAI 的 disableListModels 默认为 true，所以当前实际不会运行到这场
|
||
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: "openai",
|
||
providerName: "OpenAI",
|
||
providerType: "openai",
|
||
sorted: 1,
|
||
},
|
||
}));
|
||
}
|
||
}
|
||
export { OpenaiPath };
|