Skip to content
This repository has been archived by the owner on Sep 19, 2024. It is now read-only.

Commit

Permalink
feat: perplexity ask
Browse files Browse the repository at this point in the history 
demo time
  • Loading branch information
@Keyrxng
Keyrxng committed Oct 25, 2023
1 parent 9a3c7fb commit 3da8f6f
Show file tree
Hide file tree
Showing 8 changed files with 311 additions and 194 deletions.
2 changes: 2 additions & 0 deletions package.json
View file
Original file line number Diff line number Diff line change
Expand Up @@ -50,6 +50,7 @@
"exponential-backoff": "^3.1.1",
"husky": "^8.0.2",
"jimp": "^0.22.4",
"js-tiktoken": "^1.0.7",
"js-yaml": "^4.1.0",
"jsonwebtoken": "^9.0.2",
"libsodium-wrappers": "^0.7.11",
Expand All @@ -63,6 +64,7 @@
"parse5": "^7.1.2",
"prettier": "^2.7.1",
"probot": "^12.2.4",
"sentencepiece-js": "^1.1.0",
"telegraf": "^4.11.2",
"tsx": "^3.12.7",
"yaml": "^2.2.2"
Expand Down
6 changes: 2 additions & 4 deletions src/bindings/config.ts
View file
Original file line number Diff line number Diff line change
Expand Up @@ -65,9 +65,7 @@ export const loadConfig = async (context: Context): Promise<BotConfig> => {
permitBaseUrl: process.env.PERMIT_BASE_URL || permitBaseUrl,
},
unassign: {
timeRangeForMaxIssue: process.env.DEFAULT_TIME_RANGE_FOR_MAX_ISSUE
? Number(process.env.DEFAULT_TIME_RANGE_FOR_MAX_ISSUE)
: timeRangeForMaxIssue,
timeRangeForMaxIssue: process.env.DEFAULT_TIME_RANGE_FOR_MAX_ISSUE ? Number(process.env.DEFAULT_TIME_RANGE_FOR_MAX_ISSUE) : timeRangeForMaxIssue,
timeRangeForMaxIssueEnabled: process.env.DEFAULT_TIME_RANGE_FOR_MAX_ISSUE_ENABLED
? process.env.DEFAULT_TIME_RANGE_FOR_MAX_ISSUE_ENABLED == "true"
: timeRangeForMaxIssueEnabled,
Expand Down Expand Up @@ -108,7 +106,7 @@ export const loadConfig = async (context: Context): Promise<BotConfig> => {
registerWalletWithVerification: registerWalletWithVerification,
},
ask: {
apiKey: openAIKey,
apiKey: process.env.OPENAI_API_KEY || openAIKey,
tokenLimit: openAITokenLimit || 0,
},
accessControl: enableAccessControl,
Expand Down
1 change: 1 addition & 0 deletions src/configs/ubiquibot-config-default.ts
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@ export const DefaultConfig: MergedConfig = {
disableAnalytics: false,
commentIncentives: false,
registerWalletWithVerification: false,
openAIKey: process.env.OPENAI_API_KEY,
promotionComment:
"\n<h6>If you enjoy the DevPool experience, please follow <a href='https://github.com/ubiquity'>Ubiquity on GitHub</a> and star <a href='https://github.com/ubiquity/devpool-directory'>this repo</a> to show your support. It helps a lot!</h6>",
defaultLabels: [],
Expand Down
Binary file added src/declarations/tokenizer.model
View file
Binary file not shown.
293 changes: 244 additions & 49 deletions src/handlers/comment/handlers/ask.ts
View file
Original file line number Diff line number Diff line change
@@ -1,9 +1,11 @@
import { getBotContext, getLogger } from "../../../bindings";
import { Payload, StreamlinedComment, UserType } from "../../../types";
import { getAllIssueComments, getAllLinkedIssuesAndPullsInBody } from "../../../helpers";
import { CreateChatCompletionRequestMessage } from "openai/resources/chat";
import { askGPT, decideContextGPT, sysMsg } from "../../../helpers/gpt";
import { ChatCompletionMessageParam } from "openai/resources/chat";
import { askGPT, gptContextTemplate, sysMsg } from "../../../helpers/gpt";
import { ErrorDiff } from "../../../utils/helpers";
import fetch from "node-fetch";
import { SentencePieceProcessor, cleanText } from "sentencepiece-js";

/**
* @param body The question to ask
Expand All @@ -13,7 +15,6 @@ export const ask = async (body: string) => {
const logger = getLogger();

const payload = context.payload as Payload;
const sender = payload.sender.login;
const issue = payload.issue;

if (!body) {
Expand All @@ -24,34 +25,56 @@ export const ask = async (body: string) => {
return `This command can only be used on issues`;
}

const chatHistory: CreateChatCompletionRequestMessage[] = [];
let chatHistory: ChatCompletionMessageParam[] = [];
const streamlined: StreamlinedComment[] = [];
let linkedPRStreamlined: StreamlinedComment[] = [];
let linkedIssueStreamlined: StreamlinedComment[] = [];

const regex = /^\/ask\s(.+)$/;
const regex = /^\/ask\s*([\s\S]*)$/;
const matches = body.match(regex);

if (matches) {
const [, body] = matches;

// standard comments
const sp = new SentencePieceProcessor();
try {
await sp.load(process.cwd() + "/src/declarations/tokenizer.model");
await sp.loadVocabulary(process.cwd() + "/src/declarations/tokenizer.model");
} catch (err) {
console.log("====================================");
console.log("err", err);
console.log("====================================");
}

const encodee = (s: string, bos = true) => {
const bosID = sp.encodeIds("<s>")[0];
const eosID = sp.encodeIds("</s>")[0];

if (typeof s !== "string") {
throw new Error("encodee only accepts strings");
}
let t = sp.encodeIds(s);

if (bos) {
t = [bosID, ...t];
}
t = [...t, eosID];
return t;
};

const comments = await getAllIssueComments(issue.number);
// raw so we can grab the <!--- { 'UbiquityAI': 'answer' } ---> tag
const commentsRaw = await getAllIssueComments(issue.number, "raw");

if (!comments) {
logger.info(`Error getting issue comments`);
return ErrorDiff(`Error getting issue comments`);
}

// add the first comment of the issue/pull request
streamlined.push({
login: issue.user.login,
body: issue.body,
});

// add the rest
comments.forEach(async (comment, i) => {
if (comment.user.type == UserType.User || commentsRaw[i].body.includes("<!--- { 'UbiquityAI': 'answer' } --->")) {
streamlined.push({
Expand All @@ -71,49 +94,221 @@ export const ask = async (body: string) => {
linkedPRStreamlined = links.linkedPrs;
}

// let chatgpt deduce what is the most relevant context
const gptDecidedContext = await decideContextGPT(chatHistory, streamlined, linkedPRStreamlined, linkedIssueStreamlined);

if (linkedIssueStreamlined.length == 0 && linkedPRStreamlined.length == 0) {
// No external context to add
chatHistory.push(
{
role: "system",
content: sysMsg,
name: "UbiquityAI",
} as CreateChatCompletionRequestMessage,
{
role: "user",
content: body,
name: sender,
} as CreateChatCompletionRequestMessage
);
} else {
chatHistory.push(
{
role: "system",
content: sysMsg, // provide the answer template
name: "UbiquityAI",
} as CreateChatCompletionRequestMessage,
{
role: "system",
content: "Original Context: " + JSON.stringify(gptDecidedContext), // provide the context
name: "system",
} as CreateChatCompletionRequestMessage,
{
role: "user",
content: "Question: " + JSON.stringify(body), // provide the question
name: "user",
} as CreateChatCompletionRequestMessage
);
const formatChat = (chat: { role?: string; content?: string; login?: string; body?: string }[]) => {
if (chat.length === 0) return "";
let chatString = "";
chat.reduce((acc, message) => {
if (!message) return acc;
const role = acc.role || acc.login;
const content = acc.content || acc.body;

chatString += `${cleanText(role)}: ${cleanText(content)}\n\n`;

acc = {
role,
content,
};

return acc;
});
console.log("chatString", chatString);
return chatString;
};

chatHistory.push(
{
role: "system",
content: gptContextTemplate,
},
{
role: "user",
content: `This issue/Pr context: \n ${JSON.stringify(streamlined)}`,
}
);

if (linkedIssueStreamlined.length > 0) {
chatHistory.push({
role: "user",
content: `Linked issue(s) context: \n ${JSON.stringify(linkedIssueStreamlined)}`,
});
} else if (linkedPRStreamlined.length > 0) {
chatHistory.push({
role: "user",
content: `Linked Pr(s) context: \n ${JSON.stringify(linkedPRStreamlined)}`,
});
}

const gptResponse = await askGPT(body, chatHistory);
const gptDecidedContext = await askGPT("ContextCall", chatHistory);

const gptAnswer = typeof gptDecidedContext === "string" ? gptDecidedContext : gptDecidedContext.answer || "";
const contextTokens = encodee(cleanText(gptAnswer));

// console.log("gptDecidedContext", gptDecidedContext);
// console.log("contextTokens", contextTokens);

// const commentBeforeQuestion = streamlined[streamlined.length - 2];
// const secondLast = streamlined[streamlined.length - 3];

// const latestComments = [commentBeforeQuestion, secondLast];

// const fmLatestComments = formatChat(latestComments);

// const fmStreamlined = formatChat(streamlined);

// const formats = [
// {
// quarter: "1st",
// content: `IssueSpec: + ${issue.body} \n LastTwoComments: ${fmLatestComments} \n Question: ${body}`,
// current: contextTokens,
// },
// {
// quarter: "2nd",
// content: `IssueSpec: + ${issue.body} \n LinkedIssueContext: ${formatChat(
// linkedIssueStreamlined
// )}\n LastTwoComments: ${fmLatestComments} \n Question: ${body}`,
// current: contextTokens,
// },
// {
// quarter: "3rd",
// content: `IssueSpec: + ${issue.body} \n LinkedIssueContext: ${formatChat(linkedIssueStreamlined)}\n LinkedPRContext: ${formatChat(
// linkedPRStreamlined
// )} LastTwoComments: ${latestComments} \n Question: ${body}`,
// current: contextTokens,
// },
// {
// quarter: "4th",
// content: `IssueSpec: + ${issue.body} \n LinkedIssueContext: ${formatChat(linkedIssueStreamlined)}\n LinkedPRContext: ${formatChat(
// linkedPRStreamlined
// )} CurrentIssueComments: ${fmStreamlined} \n Question: ${body}`,
// current: contextTokens,
// },
// ];

// const remainingTokens = (s: string) => {
// const max = 4096;
// const tokens = encodee(s).length;
// let nextActiveFourth = "";

// if (tokens < max / 4) {
// nextActiveFourth = "1st";
// } else if (tokens < max / 2) {
// nextActiveFourth = "2nd";
// } else if (tokens < (max / 4) * 3) {
// nextActiveFourth = "3rd";
// } else if (tokens < max) {
// nextActiveFourth = "4th";
// } else {
// nextActiveFourth = "Max";
// }

// const remaining = {
// content: s,
// current: tokens + formats[0].current,
// quarter: nextActiveFourth,
// };

// return remaining;
// };

// let selectedFormat = "";
// let closestTokenCount = 0;

// for (const format of formats) {
// const tokenCount = remainingTokens(cleanText(format.content)).current;
// if (tokenCount > closestTokenCount && tokenCount < 4096) {
// closestTokenCount = tokenCount;
// selectedFormat = format.content;
// }
// }

// if (selectedFormat === "") {
// return "Format selection failed.";
// }

// console.log("========================");
// console.log("=== selectedFormat ===", selectedFormat);

// const { quarter, current } = remainingTokens(cleanText(selectedFormat));

// if (current >= 4096) {
// return "Format selection failed.";
// }

chatHistory = [];

const tokenSize = contextTokens.length + encodee(body).length;

if (tokenSize > 4096) {
return "Your question is too long. Please ask a shorter question.";
}

chatHistory.push(
{
role: "system",
content: `${sysMsg}`,
},
{
role: "user",
content: `Context: ${cleanText(gptAnswer)} \n Question: ${body}`,
}
);

const chats = chatHistory.map((chat) => {
return {
role: chat.role,
content: chat.content ? cleanText(chat.content) : "",
};
});

const finalTokens = encodee(formatChat(chats), false);

const options = {
method: "POST",
headers: {
accept: "application/json",
"content-type": "application/json",
authorization: "Bearer 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",
},
body: JSON.stringify({
model: "mistral-7b-instruct",
messages: chatHistory,
}),
};

const ans = await fetch("https://api.perplexity.ai/chat/completions", options).then((response) => response.json().catch((err) => console.log(err)));
const answer = { tokens: ans.usage, text: ans.choices[0].message.content };
const gptRes = await askGPT(body, chatHistory);

const gptAns = typeof gptRes === "string" ? gptRes : gptRes.answer || "";
const gptTokens = typeof gptRes === "string" ? [] : gptRes.tokenUsage || [];

const comment = `
### Perp Tokens
\`\`\`json
${JSON.stringify(answer.tokens)}
\`\`\`
### GPT Tokens
\`\`\`json
${JSON.stringify(gptTokens)}
\`\`\
### SPP Tokens
\`\`\`json
Note: JSON in responses are throwing this off rn: ${finalTokens.length + contextTokens.length} tokens
\`\`\`
### Perp Response
${answer.text}
</hr>
### GPT Response
${gptAns}
`;

if (typeof gptResponse === "string") {
return gptResponse;
} else if (gptResponse.answer) {
return gptResponse.answer;
if (answer) {
return comment;
} else {
return ErrorDiff(`Error getting response from GPT`);
}
Expand Down
Loading

0 comments on commit 3da8f6f

Please sign in to comment.