commit
affc194cde
|
@ -160,6 +160,7 @@ export class ChatGPTApi implements LLMApi {
|
|||
let requestPayload: RequestPayload | DalleRequestPayload;
|
||||
|
||||
const isDalle3 = _isDalle3(options.config.model);
|
||||
const isO1 = options.config.model.startsWith("o1");
|
||||
if (isDalle3) {
|
||||
const prompt = getMessageTextContent(
|
||||
options.messages.slice(-1)?.pop() as any,
|
||||
|
@ -181,17 +182,19 @@ export class ChatGPTApi implements LLMApi {
|
|||
const content = visionModel
|
||||
? await preProcessImageContent(v.content)
|
||||
: getMessageTextContent(v);
|
||||
messages.push({ role: v.role, content });
|
||||
if (!(isO1 && v.role === "system"))
|
||||
messages.push({ role: v.role, content });
|
||||
}
|
||||
|
||||
// O1 not support image, tools (plugin in ChatGPTNextWeb) and system, stream, logprobs, temperature, top_p, n, presence_penalty, frequency_penalty yet.
|
||||
requestPayload = {
|
||||
messages,
|
||||
stream: options.config.stream,
|
||||
stream: !isO1 ? options.config.stream : false,
|
||||
model: modelConfig.model,
|
||||
temperature: modelConfig.temperature,
|
||||
presence_penalty: modelConfig.presence_penalty,
|
||||
frequency_penalty: modelConfig.frequency_penalty,
|
||||
top_p: modelConfig.top_p,
|
||||
temperature: !isO1 ? modelConfig.temperature : 1,
|
||||
presence_penalty: !isO1 ? modelConfig.presence_penalty : 0,
|
||||
frequency_penalty: !isO1 ? modelConfig.frequency_penalty : 0,
|
||||
top_p: !isO1 ? modelConfig.top_p : 1,
|
||||
// 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.
|
||||
};
|
||||
|
@ -204,7 +207,7 @@ export class ChatGPTApi implements LLMApi {
|
|||
|
||||
console.log("[Request] openai payload: ", requestPayload);
|
||||
|
||||
const shouldStream = !isDalle3 && !!options.config.stream;
|
||||
const shouldStream = !isDalle3 && !!options.config.stream && !isO1;
|
||||
const controller = new AbortController();
|
||||
options.onController?.(controller);
|
||||
|
||||
|
@ -313,7 +316,7 @@ export class ChatGPTApi implements LLMApi {
|
|||
// make a fetch request
|
||||
const requestTimeoutId = setTimeout(
|
||||
() => controller.abort(),
|
||||
isDalle3 ? REQUEST_TIMEOUT_MS * 2 : REQUEST_TIMEOUT_MS, // dalle3 using b64_json is slow.
|
||||
isDalle3 || isO1 ? REQUEST_TIMEOUT_MS * 2 : REQUEST_TIMEOUT_MS, // dalle3 using b64_json is slow.
|
||||
);
|
||||
|
||||
const res = await fetch(chatPath, chatPayload);
|
||||
|
|
|
@ -250,6 +250,8 @@ export const KnowledgeCutOffDate: Record<string, string> = {
|
|||
"gpt-4o-mini": "2023-10",
|
||||
"gpt-4o-mini-2024-07-18": "2023-10",
|
||||
"gpt-4-vision-preview": "2023-04",
|
||||
"o1-mini": "2023-10",
|
||||
"o1-preview": "2023-10",
|
||||
// After improvements,
|
||||
// it's now easier to add "KnowledgeCutOffDate" instead of stupid hardcoding it, as was done previously.
|
||||
"gemini-pro": "2023-12",
|
||||
|
@ -276,6 +278,8 @@ const openaiModels = [
|
|||
"gpt-4-turbo-2024-04-09",
|
||||
"gpt-4-1106-preview",
|
||||
"dall-e-3",
|
||||
"o1-mini",
|
||||
"o1-preview"
|
||||
];
|
||||
|
||||
const googleModels = [
|
||||
|
|
Loading…
Reference in New Issue