🚀 Feat: Streamline File Strategies & GPT-4-Vision Settings (#1535)

* chore: fix `endpoint` typescript issues and typo in console info message

* feat(api): files GET endpoint and save only file_id references to messages

* refactor(client): `useGetFiles` query hook, update file types, optimistic update of filesQuery on file upload

* refactor(buildTree): update to use params object and accept fileMap

* feat: map files to messages; refactor(ChatView): messages only available after files are fetched

* fix: fetch files only when authenticated

* feat(api): AppService
- rename app.locals.configs to app.locals.paths
- load custom config use fileStrategy from yaml config in app.locals

* refactor: separate Firebase and Local strategies, call based on config

* refactor: modularize file strategies and employ with use of DALL-E

* refactor(librechat.yaml): add fileStrategy field

* feat: add source to MongoFile schema, as well as BatchFile, and ExtendedFile types

* feat: employ file strategies for upload/delete files

* refactor(deleteFirebaseFile): add user id validation for firebase file deletion

* chore(deleteFirebaseFile): update jsdocs

* feat: employ strategies for vision requests

* fix(client): handle messages with deleted files

* fix(client): ensure `filesToDelete` always saves/sends `file.source`

* feat(openAI): configurable `resendImages` and `imageDetail`

* refactor(getTokenCountForMessage): recursive process only when array of Objects and only their values (not keys) aside from `image_url` types

* feat(OpenAIClient): calculateImageTokenCost

* chore: remove comment

* refactor(uploadAvatar): employ fileStrategy for avatars, from social logins or user upload

* docs: update docs on how to configure fileStrategy

* fix(ci): mock winston and winston related modules, update DALLE3.spec.js with changes made

* refactor(redis): change terminal message to reflect current development state

* fix(DALL-E-2): pass fileStrategy to dall-e
This commit is contained in:
Danny Avila 2024-01-11 11:37:54 -05:00 committed by GitHub
parent 28a6807176
commit d20970f5c5
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
81 changed files with 1729 additions and 855 deletions

View file

@ -0,0 +1,78 @@
const sharp = require('sharp');
const fs = require('fs').promises;
const fetch = require('node-fetch');
const User = require('~/models/User');
const { getStrategyFunctions } = require('~/server/services/Files/strategies');
const { logger } = require('~/config');
async function convertToWebP(inputBuffer) {
return sharp(inputBuffer).resize({ width: 150 }).toFormat('webp').toBuffer();
}
/**
* Uploads an avatar image for a user. This function can handle various types of input (URL, Buffer, or File object),
* processes the image to a square format, converts it to WebP format, and then uses a specified file strategy for
* further processing. It performs validation on the user ID and the input type. The function can throw errors for
* invalid input types, fetching issues, or other processing errors.
*
* @param {Object} params - The parameters object.
* @param {string} params.userId - The unique identifier of the user for whom the avatar is being uploaded.
* @param {FileSources} params.fileStrategy - The file handling strategy to use, determining how the avatar is processed.
* @param {(string|Buffer|File)} params.input - The input representing the avatar image. Can be a URL (string),
* a Buffer, or a File object.
* @param {string} params.manual - A string flag indicating whether the upload process is manual.
*
* @returns {Promise<any>}
* A promise that resolves to the result of the `processAvatar` function, specific to the chosen file
* strategy. Throws an error if any step in the process fails.
*
* @throws {Error} Throws an error if the user ID is undefined, the input type is invalid, the image fetching fails,
* or any other error occurs during the processing.
*/
async function uploadAvatar({ userId, fileStrategy, input, manual }) {
try {
if (userId === undefined) {
throw new Error('User ID is undefined');
}
const _id = userId;
// TODO: remove direct use of Model, `User`
const oldUser = await User.findOne({ _id });
let imageBuffer;
if (typeof input === 'string') {
const response = await fetch(input);
if (!response.ok) {
throw new Error(`Failed to fetch image from URL. Status: ${response.status}`);
}
imageBuffer = await response.buffer();
} else if (input instanceof Buffer) {
imageBuffer = input;
} else if (typeof input === 'object' && input instanceof File) {
const fileContent = await fs.readFile(input.path);
imageBuffer = Buffer.from(fileContent);
} else {
throw new Error('Invalid input type. Expected URL, Buffer, or File.');
}
const { width, height } = await sharp(imageBuffer).metadata();
const minSize = Math.min(width, height);
const squaredBuffer = await sharp(imageBuffer)
.extract({
left: Math.floor((width - minSize) / 2),
top: Math.floor((height - minSize) / 2),
width: minSize,
height: minSize,
})
.toBuffer();
const webPBuffer = await convertToWebP(squaredBuffer);
const { processAvatar } = getStrategyFunctions(fileStrategy);
return await processAvatar({ buffer: webPBuffer, User: oldUser, manual });
} catch (error) {
logger.error('Error uploading the avatar:', error);
throw error;
}
}
module.exports = uploadAvatar;

View file

@ -1,29 +0,0 @@
const { ref, uploadBytes, getDownloadURL } = require('firebase/storage');
const { getFirebaseStorage } = require('~/server/services/Files/Firebase/initialize');
const { logger } = require('~/config');
async function firebaseStrategy(userId, webPBuffer, oldUser, manual) {
try {
const storage = getFirebaseStorage();
if (!storage) {
throw new Error('Firebase is not initialized.');
}
const avatarRef = ref(storage, `images/${userId.toString()}/avatar`);
await uploadBytes(avatarRef, webPBuffer);
const urlFirebase = await getDownloadURL(avatarRef);
const isManual = manual === 'true';
const url = `${urlFirebase}?manual=${isManual}`;
if (isManual) {
oldUser.avatar = url;
await oldUser.save();
}
return url;
} catch (error) {
logger.error('Error uploading profile picture:', error);
throw error;
}
}
module.exports = firebaseStrategy;

View file

@ -1,32 +0,0 @@
const fs = require('fs').promises;
const path = require('path');
async function localStrategy(userId, webPBuffer, oldUser, manual) {
const userDir = path.resolve(
__dirname,
'..',
'..',
'..',
'..',
'..',
'..',
'client',
'public',
'images',
userId,
);
let avatarPath = path.join(userDir, 'avatar.png');
const urlRoute = `/images/${userId}/avatar.png`;
await fs.mkdir(userDir, { recursive: true });
await fs.writeFile(avatarPath, webPBuffer);
const isManual = manual === 'true';
let url = `${urlRoute}?manual=${isManual}&timestamp=${new Date().getTime()}`;
if (isManual) {
oldUser.avatar = url;
await oldUser.save();
}
return url;
}
module.exports = localStrategy;

View file

@ -1,63 +0,0 @@
const sharp = require('sharp');
const fetch = require('node-fetch');
const fs = require('fs').promises;
const User = require('~/models/User');
const { getFirebaseStorage } = require('~/server/services/Files/Firebase/initialize');
const firebaseStrategy = require('./firebaseStrategy');
const localStrategy = require('./localStrategy');
const { logger } = require('~/config');
async function convertToWebP(inputBuffer) {
return sharp(inputBuffer).resize({ width: 150 }).toFormat('webp').toBuffer();
}
async function uploadAvatar(userId, input, manual) {
try {
if (userId === undefined) {
throw new Error('User ID is undefined');
}
const _id = userId;
// TODO: remove direct use of Model, `User`
const oldUser = await User.findOne({ _id });
let imageBuffer;
if (typeof input === 'string') {
const response = await fetch(input);
if (!response.ok) {
throw new Error(`Failed to fetch image from URL. Status: ${response.status}`);
}
imageBuffer = await response.buffer();
} else if (input instanceof Buffer) {
imageBuffer = input;
} else if (typeof input === 'object' && input instanceof File) {
const fileContent = await fs.readFile(input.path);
imageBuffer = Buffer.from(fileContent);
} else {
throw new Error('Invalid input type. Expected URL, Buffer, or File.');
}
const { width, height } = await sharp(imageBuffer).metadata();
const minSize = Math.min(width, height);
const squaredBuffer = await sharp(imageBuffer)
.extract({
left: Math.floor((width - minSize) / 2),
top: Math.floor((height - minSize) / 2),
width: minSize,
height: minSize,
})
.toBuffer();
const webPBuffer = await convertToWebP(squaredBuffer);
const storage = getFirebaseStorage();
if (storage) {
const url = await firebaseStrategy(userId, webPBuffer, oldUser, manual);
return url;
}
const url = await localStrategy(userId, webPBuffer, oldUser, manual);
return url;
} catch (error) {
logger.error('Error uploading the avatar:', error);
throw error;
}
}
module.exports = uploadAvatar;

View file

@ -1,36 +0,0 @@
const path = require('path');
const sharp = require('sharp');
const fs = require('fs');
const { resizeImage } = require('./resize');
async function convertToWebP(req, file, resolution = 'high') {
const inputFilePath = file.path;
const { buffer: resizedBuffer, width, height } = await resizeImage(inputFilePath, resolution);
const extension = path.extname(inputFilePath);
const { imageOutput } = req.app.locals.config;
const userPath = path.join(imageOutput, req.user.id);
if (!fs.existsSync(userPath)) {
fs.mkdirSync(userPath, { recursive: true });
}
const newPath = path.join(userPath, path.basename(inputFilePath));
if (extension.toLowerCase() === '.webp') {
const bytes = Buffer.byteLength(resizedBuffer);
await fs.promises.writeFile(newPath, resizedBuffer);
const filepath = path.posix.join('/', 'images', req.user.id, path.basename(newPath));
return { filepath, bytes, width, height };
}
const outputFilePath = newPath.replace(extension, '.webp');
const data = await sharp(resizedBuffer).toFormat('webp').toBuffer();
await fs.promises.writeFile(outputFilePath, data);
const bytes = Buffer.byteLength(data);
const filepath = path.posix.join('/', 'images', req.user.id, path.basename(outputFilePath));
await fs.promises.unlink(inputFilePath);
return { filepath, bytes, width, height };
}
module.exports = { convertToWebP };

View file

@ -1,45 +1,5 @@
const fs = require('fs');
const path = require('path');
const { EModelEndpoint } = require('librechat-data-provider');
const { updateFile } = require('~/models');
/**
* Encodes an image file to base64.
* @param {string} imagePath - The path to the image file.
* @returns {Promise<string>} A promise that resolves with the base64 encoded image data.
*/
function encodeImage(imagePath) {
return new Promise((resolve, reject) => {
fs.readFile(imagePath, (err, data) => {
if (err) {
reject(err);
} else {
resolve(data.toString('base64'));
}
});
});
}
/**
* Updates the file and encodes the image.
* @param {Object} req - The request object.
* @param {Object} file - The file object.
* @returns {Promise<[MongoFile, string]>} - A promise that resolves to an array of results from updateFile and encodeImage.
*/
async function updateAndEncode(req, file) {
const { publicPath, imageOutput } = req.app.locals.config;
const userPath = path.join(imageOutput, req.user.id);
if (!fs.existsSync(userPath)) {
fs.mkdirSync(userPath, { recursive: true });
}
const filepath = path.join(publicPath, file.filepath);
const promises = [];
promises.push(updateFile({ file_id: file.file_id }));
promises.push(encodeImage(filepath));
return await Promise.all(promises);
}
const { EModelEndpoint, FileSources } = require('librechat-data-provider');
const { getStrategyFunctions } = require('../strategies');
/**
* Encodes and formats the given files.
@ -50,25 +10,42 @@ async function updateAndEncode(req, file) {
*/
async function encodeAndFormat(req, files, endpoint) {
const promises = [];
const encodingMethods = {};
for (let file of files) {
promises.push(updateAndEncode(req, file));
const source = file.source ?? FileSources.local;
if (encodingMethods[source]) {
promises.push(encodingMethods[source](req, file));
continue;
}
const { prepareImagePayload } = getStrategyFunctions(source);
if (!prepareImagePayload) {
throw new Error(`Encoding function not implemented for ${source}`);
}
encodingMethods[source] = prepareImagePayload;
promises.push(prepareImagePayload(req, file));
}
// TODO: make detail configurable, as of now resizing is done
// to prefer "high" but "low" may be used if the image is small enough
const detail = req.body.detail ?? 'auto';
const encodedImages = await Promise.all(promises);
const detail = req.body.imageDetail ?? 'auto';
/** @type {Array<[MongoFile, string]>} */
const formattedImages = await Promise.all(promises);
const result = {
files: [],
image_urls: [],
};
for (const [file, base64] of encodedImages) {
for (const [file, imageContent] of formattedImages) {
const imagePart = {
type: 'image_url',
image_url: {
url: `data:image/webp;base64,${base64}`,
url: imageContent.startsWith('http')
? imageContent
: `data:image/webp;base64,${imageContent}`,
detail,
},
};
@ -81,17 +58,16 @@ async function encodeAndFormat(req, files, endpoint) {
result.files.push({
file_id: file.file_id,
filepath: file.filepath,
filename: file.filename,
type: file.type,
height: file.height,
width: file.width,
// filepath: file.filepath,
// filename: file.filename,
// type: file.type,
// height: file.height,
// width: file.width,
});
}
return result;
}
module.exports = {
encodeImage,
encodeAndFormat,
};

View file

@ -1,15 +1,13 @@
const convert = require('./convert');
const avatar = require('./avatar');
const encode = require('./encode');
const parse = require('./parse');
const resize = require('./resize');
const validate = require('./validate');
const uploadAvatar = require('./avatar/uploadAvatar');
module.exports = {
...convert,
...encode,
...parse,
...resize,
...validate,
uploadAvatar,
avatar,
};

View file

@ -1,7 +1,7 @@
const URL = require('url').URL;
const path = require('path');
const imageExtensionRegex = /\.(jpg|jpeg|png|gif|bmp|tiff|svg)$/i;
const imageExtensionRegex = /\.(jpg|jpeg|png|gif|bmp|tiff|svg|webp)$/i;
/**
* Extracts the image basename from a given URL.
@ -22,6 +22,24 @@ function getImageBasename(urlString) {
}
}
/**
* Extracts the basename of a file from a given URL.
*
* @param {string} urlString - The URL string from which the file basename is to be extracted.
* @returns {string} The basename of the file from the URL.
* Returns an empty string if the URL parsing fails.
*/
function getFileBasename(urlString) {
try {
const url = new URL(urlString);
return path.basename(url.pathname);
} catch (error) {
// If URL parsing fails, return an empty string
return '';
}
}
module.exports = {
getImageBasename,
getFileBasename,
};