diff --git a/Frontend/package-lock.json b/Frontend/package-lock.json index 5ac2281..c6e36bd 100644 --- a/Frontend/package-lock.json +++ b/Frontend/package-lock.json @@ -8,7 +8,6 @@ "name": "my-react-app", "version": "0.1.0", "dependencies": { - "@mui/material": "^7.1.1", "@testing-library/dom": "^10.4.0", "@testing-library/jest-dom": "^6.6.3", "@testing-library/react": "^16.2.0", @@ -2356,68 +2355,6 @@ "postcss-selector-parser": "^6.0.10" } }, - "node_modules/@emotion/cache": { - "version": "11.14.0", - "resolved": "https://registry.npmjs.org/@emotion/cache/-/cache-11.14.0.tgz", - "integrity": "sha512-L/B1lc/TViYk4DcpGxtAVbx0ZyiKM5ktoIyafGkH6zg/tj+mA+NE//aPYKG0k8kCHSHVJrpLpcAlOBEXQ3SavA==", - "license": "MIT", - "dependencies": { - "@emotion/memoize": "^0.9.0", - "@emotion/sheet": "^1.4.0", - "@emotion/utils": "^1.4.2", - "@emotion/weak-memoize": "^0.4.0", - "stylis": "4.2.0" - } - }, - "node_modules/@emotion/hash": { - "version": "0.9.2", - "resolved": "https://registry.npmjs.org/@emotion/hash/-/hash-0.9.2.tgz", - "integrity": "sha512-MyqliTZGuOm3+5ZRSaaBGP3USLw6+EGykkwZns2EPC5g8jJ4z9OrdZY9apkl3+UP9+sdz76YYkwCKP5gh8iY3g==", - "license": "MIT" - }, - "node_modules/@emotion/memoize": { - "version": "0.9.0", - "resolved": "https://registry.npmjs.org/@emotion/memoize/-/memoize-0.9.0.tgz", - "integrity": "sha512-30FAj7/EoJ5mwVPOWhAyCX+FPfMDrVecJAM+Iw9NRoSl4BBAQeqj4cApHHUXOVvIPgLVDsCFoz/hGD+5QQD1GQ==", - "license": "MIT" - }, - "node_modules/@emotion/serialize": { - "version": "1.3.3", - "resolved": "https://registry.npmjs.org/@emotion/serialize/-/serialize-1.3.3.tgz", - "integrity": "sha512-EISGqt7sSNWHGI76hC7x1CksiXPahbxEOrC5RjmFRJTqLyEK9/9hZvBbiYn70dw4wuwMKiEMCUlR6ZXTSWQqxA==", - "license": "MIT", - "dependencies": { - "@emotion/hash": "^0.9.2", - "@emotion/memoize": "^0.9.0", - "@emotion/unitless": "^0.10.0", - "@emotion/utils": "^1.4.2", - "csstype": "^3.0.2" - } - }, - "node_modules/@emotion/sheet": { - "version": "1.4.0", - "resolved": "https://registry.npmjs.org/@emotion/sheet/-/sheet-1.4.0.tgz", - "integrity": "sha512-fTBW9/8r2w3dXWYM4HCB1Rdp8NLibOw2+XELH5m5+AkWiL/KqYX6dc0kKYlaYyKjrQ6ds33MCdMPEwgs2z1rqg==", - "license": "MIT" - }, - "node_modules/@emotion/unitless": { - "version": "0.10.0", - "resolved": "https://registry.npmjs.org/@emotion/unitless/-/unitless-0.10.0.tgz", - "integrity": "sha512-dFoMUuQA20zvtVTuxZww6OHoJYgrzfKM1t52mVySDJnMSEa08ruEvdYQbhvyu6soU+NeLVd3yKfTfT0NeV6qGg==", - "license": "MIT" - }, - "node_modules/@emotion/utils": { - "version": "1.4.2", - "resolved": "https://registry.npmjs.org/@emotion/utils/-/utils-1.4.2.tgz", - "integrity": "sha512-3vLclRofFziIa3J2wDh9jjbkUz9qk5Vi3IZ/FSTKViB0k+ef0fPV7dYrUIugbgupYDx7v9ud/SjrtEP8Y4xLoA==", - "license": "MIT" - }, - "node_modules/@emotion/weak-memoize": { - "version": "0.4.0", - "resolved": "https://registry.npmjs.org/@emotion/weak-memoize/-/weak-memoize-0.4.0.tgz", - "integrity": "sha512-snKqtPW01tN0ui7yu9rGv69aJXr/a/Ywvl11sUjNtEcRc+ng/mQriFL0wLXMef74iHa/EkftbDzU9F8iFbH+zg==", - "license": "MIT" - }, "node_modules/@eslint-community/eslint-utils": { "version": "4.7.0", "resolved": "https://registry.npmjs.org/@eslint-community/eslint-utils/-/eslint-utils-4.7.0.tgz", @@ -3030,213 +2967,6 @@ "integrity": "sha512-Vo+PSpZG2/fmgmiNzYK9qWRh8h/CHrwD0mo1h1DzL4yzHNSfWYujGTYsWGreD000gcgmZ7K4Ys6Tx9TxtsKdDw==", "license": "MIT" }, - "node_modules/@mui/core-downloads-tracker": { - "version": "7.1.1", - "resolved": "https://registry.npmjs.org/@mui/core-downloads-tracker/-/core-downloads-tracker-7.1.1.tgz", - "integrity": "sha512-yBckQs4aQ8mqukLnPC6ivIRv6guhaXi8snVl00VtyojBbm+l6VbVhyTSZ68Abcx7Ah8B+GZhrB7BOli+e+9LkQ==", - "license": "MIT", - "funding": { - "type": "opencollective", - "url": "https://opencollective.com/mui-org" - } - }, - "node_modules/@mui/material": { - "version": "7.1.1", - "resolved": "https://registry.npmjs.org/@mui/material/-/material-7.1.1.tgz", - "integrity": "sha512-mTpdmdZCaHCGOH3SrYM41+XKvNL0iQfM9KlYgpSjgadXx/fEKhhvOktxm8++Xw6FFeOHoOiV+lzOI8X1rsv71A==", - "license": "MIT", - "dependencies": { - "@babel/runtime": "^7.27.1", - "@mui/core-downloads-tracker": "^7.1.1", - "@mui/system": "^7.1.1", - "@mui/types": "^7.4.3", - "@mui/utils": "^7.1.1", - "@popperjs/core": "^2.11.8", - "@types/react-transition-group": "^4.4.12", - "clsx": "^2.1.1", - "csstype": "^3.1.3", - "prop-types": "^15.8.1", - "react-is": "^19.1.0", - "react-transition-group": "^4.4.5" - }, - "engines": { - "node": ">=14.0.0" - }, - "funding": { - "type": "opencollective", - "url": "https://opencollective.com/mui-org" - }, - "peerDependencies": { - "@emotion/react": "^11.5.0", - "@emotion/styled": "^11.3.0", - "@mui/material-pigment-css": "^7.1.1", - "@types/react": "^17.0.0 || ^18.0.0 || ^19.0.0", - "react": "^17.0.0 || ^18.0.0 || ^19.0.0", - "react-dom": "^17.0.0 || ^18.0.0 || ^19.0.0" - }, - "peerDependenciesMeta": { - "@emotion/react": { - "optional": true - }, - "@emotion/styled": { - "optional": true - }, - "@mui/material-pigment-css": { - "optional": true - }, - "@types/react": { - "optional": true - } - } - }, - "node_modules/@mui/private-theming": { - "version": "7.1.1", - "resolved": "https://registry.npmjs.org/@mui/private-theming/-/private-theming-7.1.1.tgz", - "integrity": "sha512-M8NbLUx+armk2ZuaxBkkMk11ultnWmrPlN0Xe3jUEaBChg/mcxa5HWIWS1EE4DF36WRACaAHVAvyekWlDQf0PQ==", - "license": "MIT", - "dependencies": { - "@babel/runtime": "^7.27.1", - "@mui/utils": "^7.1.1", - "prop-types": "^15.8.1" - }, - "engines": { - "node": ">=14.0.0" - }, - "funding": { - "type": "opencollective", - "url": "https://opencollective.com/mui-org" - }, - "peerDependencies": { - "@types/react": "^17.0.0 || ^18.0.0 || ^19.0.0", - "react": "^17.0.0 || ^18.0.0 || ^19.0.0" - }, - "peerDependenciesMeta": { - "@types/react": { - "optional": true - } - } - }, - "node_modules/@mui/styled-engine": { - "version": "7.1.1", - "resolved": "https://registry.npmjs.org/@mui/styled-engine/-/styled-engine-7.1.1.tgz", - "integrity": "sha512-R2wpzmSN127j26HrCPYVQ53vvMcT5DaKLoWkrfwUYq3cYytL6TQrCH8JBH3z79B6g4nMZZVoaXrxO757AlShaw==", - "license": "MIT", - "dependencies": { - "@babel/runtime": "^7.27.1", - "@emotion/cache": "^11.13.5", - "@emotion/serialize": "^1.3.3", - "@emotion/sheet": "^1.4.0", - "csstype": "^3.1.3", - "prop-types": "^15.8.1" - }, - "engines": { - "node": ">=14.0.0" - }, - "funding": { - "type": "opencollective", - "url": "https://opencollective.com/mui-org" - }, - "peerDependencies": { - "@emotion/react": "^11.4.1", - "@emotion/styled": "^11.3.0", - "react": "^17.0.0 || ^18.0.0 || ^19.0.0" - }, - "peerDependenciesMeta": { - "@emotion/react": { - "optional": true - }, - "@emotion/styled": { - "optional": true - } - } - }, - "node_modules/@mui/system": { - "version": "7.1.1", - "resolved": "https://registry.npmjs.org/@mui/system/-/system-7.1.1.tgz", - "integrity": "sha512-Kj1uhiqnj4Zo7PDjAOghtXJtNABunWvhcRU0O7RQJ7WOxeynoH6wXPcilphV8QTFtkKaip8EiNJRiCD+B3eROA==", - "license": "MIT", - "dependencies": { - "@babel/runtime": "^7.27.1", - "@mui/private-theming": "^7.1.1", - "@mui/styled-engine": "^7.1.1", - "@mui/types": "^7.4.3", - "@mui/utils": "^7.1.1", - "clsx": "^2.1.1", - "csstype": "^3.1.3", - "prop-types": "^15.8.1" - }, - "engines": { - "node": ">=14.0.0" - }, - "funding": { - "type": "opencollective", - "url": "https://opencollective.com/mui-org" - }, - "peerDependencies": { - "@emotion/react": "^11.5.0", - "@emotion/styled": "^11.3.0", - "@types/react": "^17.0.0 || ^18.0.0 || ^19.0.0", - "react": "^17.0.0 || ^18.0.0 || ^19.0.0" - }, - "peerDependenciesMeta": { - "@emotion/react": { - "optional": true - }, - "@emotion/styled": { - "optional": true - }, - "@types/react": { - "optional": true - } - } - }, - "node_modules/@mui/types": { - "version": "7.4.3", - "resolved": "https://registry.npmjs.org/@mui/types/-/types-7.4.3.tgz", - "integrity": "sha512-2UCEiK29vtiZTeLdS2d4GndBKacVyxGvReznGXGr+CzW/YhjIX+OHUdCIczZjzcRAgKBGmE9zCIgoV9FleuyRQ==", - "license": "MIT", - "dependencies": { - "@babel/runtime": "^7.27.1" - }, - "peerDependencies": { - "@types/react": "^17.0.0 || ^18.0.0 || ^19.0.0" - }, - "peerDependenciesMeta": { - "@types/react": { - "optional": true - } - } - }, - "node_modules/@mui/utils": { - "version": "7.1.1", - "resolved": "https://registry.npmjs.org/@mui/utils/-/utils-7.1.1.tgz", - "integrity": "sha512-BkOt2q7MBYl7pweY2JWwfrlahhp+uGLR8S+EhiyRaofeRYUWL2YKbSGQvN4hgSN1i8poN0PaUiii1kEMrchvzg==", - "license": "MIT", - "dependencies": { - "@babel/runtime": "^7.27.1", - "@mui/types": "^7.4.3", - "@types/prop-types": "^15.7.14", - "clsx": "^2.1.1", - "prop-types": "^15.8.1", - "react-is": "^19.1.0" - }, - "engines": { - "node": ">=14.0.0" - }, - "funding": { - "type": "opencollective", - "url": "https://opencollective.com/mui-org" - }, - "peerDependencies": { - "@types/react": "^17.0.0 || ^18.0.0 || ^19.0.0", - "react": "^17.0.0 || ^18.0.0 || ^19.0.0" - }, - "peerDependenciesMeta": { - "@types/react": { - "optional": true - } - } - }, "node_modules/@nicolo-ribaudo/eslint-scope-5-internals": { "version": "5.1.1-v1", "resolved": "https://registry.npmjs.org/@nicolo-ribaudo/eslint-scope-5-internals/-/eslint-scope-5-internals-5.1.1-v1.tgz", @@ -3361,16 +3091,6 @@ } } }, - "node_modules/@popperjs/core": { - "version": "2.11.8", - "resolved": "https://registry.npmjs.org/@popperjs/core/-/core-2.11.8.tgz", - "integrity": "sha512-P1st0aksCrn9sGZhp8GMYwBnQsbvAWsZAX44oXNNvLHGqAOcoVxmjZiohstwQ7SqKnbR47akdNi+uleWD8+g6A==", - "license": "MIT", - "funding": { - "type": "opencollective", - "url": "https://opencollective.com/popperjs" - } - }, "node_modules/@rollup/plugin-babel": { "version": "5.3.1", "resolved": "https://registry.npmjs.org/@rollup/plugin-babel/-/plugin-babel-5.3.1.tgz", @@ -4087,12 +3807,6 @@ "integrity": "sha512-+68kP9yzs4LMp7VNh8gdzMSPZFL44MLGqiHWvttYJe+6qnuVr4Ek9wSBQoveqY/r+LwjCcU29kNVkidwim+kYA==", "license": "MIT" }, - "node_modules/@types/prop-types": { - "version": "15.7.15", - "resolved": "https://registry.npmjs.org/@types/prop-types/-/prop-types-15.7.15.tgz", - "integrity": "sha512-F6bEyamV9jKGAFBEmlQnesRPGOQqS2+Uwi0Em15xenOxHaf2hv6L8YCVn3rPdPJOiJfPiCnLIRyvwVaqMY3MIw==", - "license": "MIT" - }, "node_modules/@types/q": { "version": "1.5.8", "resolved": "https://registry.npmjs.org/@types/q/-/q-1.5.8.tgz", @@ -4116,20 +3830,12 @@ "resolved": "https://registry.npmjs.org/@types/react/-/react-19.1.8.tgz", "integrity": "sha512-AwAfQ2Wa5bCx9WP8nZL2uMZWod7J7/JSplxbTmBQ5ms6QpqNYm672H0Vu9ZVKVngQ+ii4R/byguVEUZQyeg44g==", "license": "MIT", + "optional": true, "peer": true, "dependencies": { "csstype": "^3.0.2" } }, - "node_modules/@types/react-transition-group": { - "version": "4.4.12", - "resolved": "https://registry.npmjs.org/@types/react-transition-group/-/react-transition-group-4.4.12.tgz", - "integrity": "sha512-8TV6R3h2j7a91c+1DXdJi3Syo69zzIZbz7Lg5tORM5LEJG7X/E6a1V3drRyBRZq7/utz7A+c4OgYLiLcYGHG6w==", - "license": "MIT", - "peerDependencies": { - "@types/react": "*" - } - }, "node_modules/@types/resolve": { "version": "1.17.1", "resolved": "https://registry.npmjs.org/@types/resolve/-/resolve-1.17.1.tgz", @@ -5934,15 +5640,6 @@ "wrap-ansi": "^7.0.0" } }, - "node_modules/clsx": { - "version": "2.1.1", - "resolved": "https://registry.npmjs.org/clsx/-/clsx-2.1.1.tgz", - "integrity": "sha512-eYm0QWBtUrBWZWG0d386OGAw16Z995PiOVo2B7bjWSbHedGl5e0ZWaq65kOGgUSNesEIDkB9ISbTg/JK9dhCZA==", - "license": "MIT", - "engines": { - "node": ">=6" - } - }, "node_modules/co": { "version": "4.6.0", "resolved": "https://registry.npmjs.org/co/-/co-4.6.0.tgz", @@ -6677,7 +6374,9 @@ "version": "3.1.3", "resolved": "https://registry.npmjs.org/csstype/-/csstype-3.1.3.tgz", "integrity": "sha512-M1uQkMl8rQK/szD0LNhtqxIPLpimGm8sOBwU7lLnCpSbTyY3yeU1Vc7l4KT5zT4s/yOxHH5O7tIuuLOCnLADRw==", - "license": "MIT" + "license": "MIT", + "optional": true, + "peer": true }, "node_modules/damerau-levenshtein": { "version": "1.0.8", @@ -7005,16 +6704,6 @@ "utila": "~0.4" } }, - "node_modules/dom-helpers": { - "version": "5.2.1", - "resolved": "https://registry.npmjs.org/dom-helpers/-/dom-helpers-5.2.1.tgz", - "integrity": "sha512-nRCa7CK3VTrM2NmGkIy4cbK7IZlgBE/PYMn55rrXefr5xXDP0LdtfPnblFDoVdcAfslJ7or6iqAUnx0CCGIWQA==", - "license": "MIT", - "dependencies": { - "@babel/runtime": "^7.8.7", - "csstype": "^3.0.2" - } - }, "node_modules/dom-serializer": { "version": "1.4.1", "resolved": "https://registry.npmjs.org/dom-serializer/-/dom-serializer-1.4.1.tgz", @@ -14268,12 +13957,6 @@ "integrity": "sha512-SN/U6Ytxf1QGkw/9ve5Y+NxBbZM6Ht95tuXNMKs8EJyFa/Vy/+Co3stop3KBHARfn/giv+Lj1uUnTfOJ3moFEQ==", "license": "MIT" }, - "node_modules/react-is": { - "version": "19.1.0", - "resolved": "https://registry.npmjs.org/react-is/-/react-is-19.1.0.tgz", - "integrity": "sha512-Oe56aUPnkHyyDxxkvqtd7KkdQP5uIUfHxd5XTb3wE9d/kRnZLmKbDB0GWk919tdQ+mxxPtG6EAs6RMT6i1qtHg==", - "license": "MIT" - }, "node_modules/react-refresh": { "version": "0.11.0", "resolved": "https://registry.npmjs.org/react-refresh/-/react-refresh-0.11.0.tgz", @@ -14394,22 +14077,6 @@ } } }, - "node_modules/react-transition-group": { - "version": "4.4.5", - "resolved": "https://registry.npmjs.org/react-transition-group/-/react-transition-group-4.4.5.tgz", - "integrity": "sha512-pZcd1MCJoiKiBR2NRxeCRg13uCXbydPnmB4EOeRrY7480qNWO8IIgQG6zlDkm6uRMsURXPuKq0GWtiM59a5Q6g==", - "license": "BSD-3-Clause", - "dependencies": { - "@babel/runtime": "^7.5.5", - "dom-helpers": "^5.0.1", - "loose-envify": "^1.4.0", - "prop-types": "^15.6.2" - }, - "peerDependencies": { - "react": ">=16.6.0", - "react-dom": ">=16.6.0" - } - }, "node_modules/read-cache": { "version": "1.0.0", "resolved": "https://registry.npmjs.org/read-cache/-/read-cache-1.0.0.tgz", @@ -15999,12 +15666,6 @@ "postcss": "^8.2.15" } }, - "node_modules/stylis": { - "version": "4.2.0", - "resolved": "https://registry.npmjs.org/stylis/-/stylis-4.2.0.tgz", - "integrity": "sha512-Orov6g6BB1sDfYgzWfTHDOxamtX1bE/zo104Dh9e6fqJ3PooipYyfJ0pUmrZO2wAvO8YbEyeFrkV91XTsGMSrw==", - "license": "MIT" - }, "node_modules/sucrase": { "version": "3.35.0", "resolved": "https://registry.npmjs.org/sucrase/-/sucrase-3.35.0.tgz", @@ -16776,9 +16437,9 @@ } }, "node_modules/typescript": { - "version": "5.8.3", - "resolved": "https://registry.npmjs.org/typescript/-/typescript-5.8.3.tgz", - "integrity": "sha512-p1diW6TqL9L07nNxvRMM7hMMw4c5XOo/1ibL4aAIGmSAt9slTE1Xgw5KWuof2uTOvCg9BY7ZRi+GaF+7sfgPeQ==", + "version": "4.9.5", + "resolved": "https://registry.npmjs.org/typescript/-/typescript-4.9.5.tgz", + "integrity": "sha512-1FXk9E2Hm+QzZQ7z+McJiHL4NW1F2EzMu9Nq9i3zAaGqibafqYwCVU6WyWAuyQRRzOlxou8xZSyXLEN8oKj24g==", "license": "Apache-2.0", "peer": true, "bin": { @@ -16786,7 +16447,7 @@ "tsserver": "bin/tsserver" }, "engines": { - "node": ">=14.17" + "node": ">=4.2.0" } }, "node_modules/unbox-primitive": { diff --git a/Frontend/src/App.js b/Frontend/src/App.js index d7e6b33..b227e82 100644 --- a/Frontend/src/App.js +++ b/Frontend/src/App.js @@ -3,12 +3,11 @@ import React, { useState, useEffect } from 'react'; import { BrowserRouter as Router, Routes, Route, useNavigate } from 'react-router-dom'; import Navbar from './components/Navbar'; import LandingPage from './pages/AppLanding'; +import ProviderSelectionPage from './pages/ProviderSelectionPage'; import DetectHardwarePage from './pages/DetectHardwarePage'; import FinetuneSettings from './pages/FinetuningSettingsPage'; import Loading from './pages/Loading'; import TechnicalDetailsPage from './pages/TechnicalDetailsPage'; -import ListModels from './pages/ListModels'; -import './index.css'; import ListAllModels from "./pages/ListAllModels"; const RedirectToFastAPI = () => { @@ -21,6 +20,7 @@ const RedirectToFastAPI = () => { function App() { const [finetuneSettings, setFinetuneSettings] = useState({ + provider: 'huggingface', // Default provider task: 'text-generation', model_name: 'llama2-7b', compute_specs: 'Standard GPU', @@ -88,10 +88,10 @@ function App() { } /> } @@ -105,6 +105,15 @@ function App() { /> } /> + + } + /> {/* } diff --git a/Frontend/src/pages/AppLanding.jsx b/Frontend/src/pages/AppLanding.jsx index 30ed26b..f181b8d 100644 --- a/Frontend/src/pages/AppLanding.jsx +++ b/Frontend/src/pages/AppLanding.jsx @@ -89,7 +89,7 @@ const LandingPage = ({ appName = "ModelForge" }) => {

- + Start Building Your AI @@ -224,7 +224,7 @@ const LandingPage = ({ appName = "ModelForge" }) => {
- + Start Your Business AI @@ -270,7 +270,7 @@ const LandingPage = ({ appName = "ModelForge" }) => { No coding. No technical skills. Just pure AI power at your fingertips.

- + Start Building Now diff --git a/Frontend/src/pages/FinetuningSettingsPage.jsx b/Frontend/src/pages/FinetuningSettingsPage.jsx index 3b7b264..d5d2aa1 100644 --- a/Frontend/src/pages/FinetuningSettingsPage.jsx +++ b/Frontend/src/pages/FinetuningSettingsPage.jsx @@ -7,7 +7,6 @@ const FinetuneSettings = ({ defaultValues, updateSettings }) => { const [showAdvanced, setShowAdvanced] = useState(false); const [selectedFile, setSelectedFile] = useState(null); const [formState, setFormState] = useState({}); - const [settingsUpdated, setSettingsUpdated] = useState(false); const [activeTooltip, setActiveTooltip] = useState(null); useEffect(() => { @@ -18,9 +17,16 @@ const FinetuneSettings = ({ defaultValues, updateSettings }) => { const data = await response.json(); console.log("Fetched default values:", data.default_values); - defaultValues = data.default_values; - // Update form state with fetched values - setFormState(data.default_values); + + // Preserve the provider from defaultValues if it exists + const preservedProvider = defaultValues?.provider || 'huggingface'; + console.log("Preserving provider:", preservedProvider); + + // Update form state with fetched values but preserve provider + setFormState({ + ...data.default_values, + provider: preservedProvider + }); } catch (err) { console.error("Error fetching settings:", err); } @@ -88,10 +94,17 @@ const FinetuneSettings = ({ defaultValues, updateSettings }) => { formData.append("json_file", selectedFile, selectedFile.name); } + // Ensure provider is included in formState + const settingsToSend = { + ...formState, + provider: formState.provider || defaultValues.provider || 'huggingface' + }; + // Append settings as a JSON string - formData.append("settings", JSON.stringify(formState)); + formData.append("settings", JSON.stringify(settingsToSend)); - console.log("Sending data to server:",JSON.stringify(formState)); + console.log("Sending data to server:", JSON.stringify(settingsToSend)); + console.log("Provider being sent:", settingsToSend.provider); try { const response = await fetch(`${config.baseURL}/finetune/load_settings`, { @@ -164,13 +177,6 @@ const FinetuneSettings = ({ defaultValues, updateSettings }) => {

Configure your model training parameters

- {settingsUpdated && ( -
-
Settings updated successfully! Redirecting...
-
-
- )} -
{/* Configuration Summary */}
@@ -192,6 +198,12 @@ const FinetuneSettings = ({ defaultValues, updateSettings }) => { {formState.model_name || defaultValues.model_name || 'Not set'}
+
+ +
+ {formState.provider || defaultValues.provider || 'huggingface'} +
+
diff --git a/Frontend/src/pages/ProviderSelectionPage.jsx b/Frontend/src/pages/ProviderSelectionPage.jsx new file mode 100644 index 0000000..d084463 --- /dev/null +++ b/Frontend/src/pages/ProviderSelectionPage.jsx @@ -0,0 +1,251 @@ +import React, { useState, useEffect } from 'react'; +import { useNavigate } from 'react-router-dom'; +import { config } from '../services/api'; + +const ProviderSelectionPage = ({ currentSettings, updateSettings }) => { + const navigate = useNavigate(); + const [providers, setProviders] = useState([]); + const [loadingProviders, setLoadingProviders] = useState(true); + const [selectedProvider, setSelectedProvider] = useState(currentSettings.provider || 'huggingface'); + const [hoveredProvider, setHoveredProvider] = useState(null); + + useEffect(() => { + const fetchProviders = async () => { + try { + const response = await fetch(`${config.baseURL}/finetune/providers`); + if (!response.ok) throw new Error('Failed to fetch providers'); + + const data = await response.json(); + console.log("Fetched providers:", data); + setProviders(data.providers || []); + setLoadingProviders(false); + } catch (err) { + console.error("Error fetching providers:", err); + setLoadingProviders(false); + } + }; + + fetchProviders(); + }, []); + + const handleProviderSelect = (providerName) => { + setSelectedProvider(providerName); + }; + + const handleContinue = () => { + // Update settings with selected provider + if (updateSettings) { + updateSettings({ provider: selectedProvider }); + } + + // Navigate to task selection (hardware detection page) + navigate('/finetune/detect'); + }; + + const providerDetails = { + huggingface: { + logo: 'https://huggingface.co/datasets/huggingface/brand-assets/resolve/main/hf-logo.svg', + features: [ + 'Standard HuggingFace transformers', + 'PEFT/LoRA fine-tuning', + '4-bit/8-bit quantization', + 'Maximum compatibility', + 'Established ecosystem' + ], + performance: 'Baseline speed & memory', + bestFor: 'General use, maximum compatibility' + }, + unsloth: { + logo: 'https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20logo%20white%20text.png', + features: [ + 'Optimized FastLanguageModel', + 'Enhanced LoRA/QLoRA', + 'Memory-efficient attention', + '~2x faster training', + '~30% less memory usage' + ], + performance: '2x faster, 30% less memory', + bestFor: 'Faster training, limited hardware', + installCommand: 'pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"' + } + }; + + return ( +
+
+ {/* Header */} +
+

+ Choose Your Fine-tuning Provider +

+

+ Select the backend that best suits your needs +

+
+ + {/* Loading State */} + {loadingProviders && ( +
+
+
+ )} + + {/* Provider Cards */} + {!loadingProviders && ( +
+ {providers.map((provider) => { + const isSelected = selectedProvider === provider.name; + const isAvailable = provider.available; + const details = providerDetails[provider.name] || {}; + + return ( +
isAvailable && handleProviderSelect(provider.name)} + onMouseEnter={() => setHoveredProvider(provider.name)} + onMouseLeave={() => setHoveredProvider(null)} + className={` + relative rounded-xl p-6 border-2 transition-all duration-300 transform + ${isAvailable ? 'cursor-pointer hover:scale-105' : 'cursor-not-allowed opacity-60'} + ${isSelected && isAvailable + ? 'border-orange-500 bg-gradient-to-br from-orange-500/20 to-transparent shadow-xl shadow-orange-500/20' + : 'border-gray-700 bg-gray-800/50 hover:border-gray-600' + } + `} + > + {/* Selected Badge */} + {isSelected && isAvailable && ( +
+ Selected +
+ )} + + {/* Not Installed Badge */} + {!isAvailable && ( +
+ Not Installed +
+ )} + + {/* Provider Header */} +
+
+ {details.logo ? ( + {`${provider.name} { + e.target.style.display = 'none'; + e.target.nextSibling.style.display = 'block'; + }} + /> + ) : null} + + + +
+
+

{provider.name}

+

{provider.description}

+
+
+ + {/* Performance Badge */} + {details.performance && ( +
+ ⚡ {details.performance} +
+ )} + + {/* Features List */} + {details.features && ( +
+

Features:

+
    + {details.features.map((feature, idx) => ( +
  • + + + + {feature} +
    • + ))} +
+
+ )} + + {/* Best For */} + {details.bestFor && ( +
+

+ Best for: {details.bestFor} +

+
+ )} + + {/* Installation Instructions */} + {!isAvailable && details.installCommand && ( +
+

⚠️ Installation Required:

+ + {details.installCommand} + +
+ )} +
+ ); + })} +
+ )} + + {/* Continue Button */} +
+ +
+ + {/* Info Footer */} +
+

You can change the provider later in the advanced settings

+
+
+
+ ); +}; + +export default ProviderSelectionPage; diff --git a/ModelForge/routers/finetuning_router.py b/ModelForge/routers/finetuning_router.py index 4161b2f..6cd6eac 100644 --- a/ModelForge/routers/finetuning_router.py +++ b/ModelForge/routers/finetuning_router.py @@ -13,6 +13,8 @@ from ..utilities.finetuning.CausalLLMTuner import CausalLLMFinetuner from ..utilities.finetuning.QuestionAnsweringTuner import QuestionAnsweringTuner from ..utilities.finetuning.Seq2SeqLMTuner import Seq2SeqFinetuner +from ..utilities.finetuning.provider_adapter import ProviderFinetuner +from ..utilities.finetuning.providers import ProviderRegistry from ..utilities.hardware_detection.model_validator import ModelValidator from ..globals.globals_instance import global_manager @@ -25,6 +27,12 @@ VALID_TASKS = ["text-generation", "summarization", "extractive-question-answering"] VALID_TASKS_STR = "'text-generation', 'summarization', or 'extractive-question-answering'" +# Valid providers (dynamically loaded from registry) +def get_valid_providers(): + """Get list of available provider names.""" + providers = ProviderRegistry.list_available() + return [p["name"] for p in providers] + ## Pydantic Data Validator Classes class TaskFormData(BaseModel): task: str @@ -53,6 +61,7 @@ def validate_repo_name(cls, repo_name): class SettingsFormData(BaseModel): task: str model_name: str + provider: str = "huggingface" # Default to huggingface for backward compatibility num_train_epochs: int compute_specs: str lora_r: int @@ -96,6 +105,15 @@ def validate_model_name(cls, model_name): if not model_name: raise ValueError("Model name cannot be empty.") return model_name + @field_validator("provider") + def validate_provider(cls, provider): + valid_providers = get_valid_providers() + if not valid_providers: + # If no providers available, default to huggingface + return "huggingface" + if provider not in valid_providers: + raise ValueError(f"Invalid provider. Must be one of {', '.join(valid_providers)}.") + return provider @field_validator("num_train_epochs") def validate_num_train_epochs(cls, num_train_epochs): if num_train_epochs <= 0: @@ -248,6 +266,31 @@ def validate_batch_size_with_compute_specs(self): return self +@router.get("/providers") +async def list_providers(request: Request) -> JSONResponse: + """ + List all available fine-tuning providers. + + Returns provider information including availability status. + """ + try: + all_providers = ProviderRegistry.list_all() + available_providers = ProviderRegistry.list_available() + + return JSONResponse({ + "status_code": 200, + "providers": all_providers, + "available": available_providers, + "default": "huggingface" + }) + except Exception as e: + print(f"Error listing providers: {e}") + raise HTTPException( + status_code=500, + detail="Error retrieving provider information" + ) + + @router.get("/detect") async def detect_hardware_page(request: Request) -> JSONResponse: global_manager.clear_settings_cache() # Clear the cache to ensure fresh detection @@ -451,15 +494,20 @@ def finetuning_task(llm_tuner) -> None: # Use the path returned from finetune (should be absolute) model_path = os.path.abspath(path) if not os.path.isabs(path) else path + # Get provider name (either from ProviderFinetuner or default to huggingface) + provider_name = getattr(llm_tuner, 'provider_name', 'huggingface') + model_data = { "model_name": global_manager.settings_builder.fine_tuned_name.split('/')[-1] if global_manager.settings_builder.fine_tuned_name else os.path.basename(model_path), "base_model": global_manager.settings_builder.model_name, "task": global_manager.settings_builder.task, "description": f"Fine-tuned {global_manager.settings_builder.model_name} for {global_manager.settings_builder.task}" + - (" (Custom Model)" if global_manager.settings_builder.is_custom_model else " (Recommended Model)"), + (" (Custom Model)" if global_manager.settings_builder.is_custom_model else " (Recommended Model)") + + f" using {provider_name}", "creation_date": datetime.now().isoformat(), "model_path": model_path, - "is_custom_base_model": global_manager.settings_builder.is_custom_model + "is_custom_base_model": global_manager.settings_builder.is_custom_model, + "provider": provider_name } global_manager.db_manager.add_model(model_data) @@ -501,6 +549,10 @@ async def start_finetuning_page(request: Request, background_task: BackgroundTas print(f"Starting finetuning with CUSTOM MODEL: {global_manager.settings_builder.model_name}") else: print(f"Starting finetuning with RECOMMENDED MODEL: {global_manager.settings_builder.model_name}") + + # Log provider selection + provider = global_manager.settings_builder.provider + print(f"Using provider: {provider}") if not global_manager.settings_cache: raise HTTPException( @@ -528,29 +580,43 @@ async def start_finetuning_page(request: Request, background_task: BackgroundTas global_manager.finetuning_status["status"] = "initializing" global_manager.finetuning_status["message"] = "Starting finetuning process..." - if global_manager.settings_builder.task == "text-generation": - llm_tuner = CausalLLMFinetuner( - model_name=global_manager.settings_builder.model_name, - compute_specs=global_manager.settings_builder.compute_profile, - pipeline_task="text-generation" - ) - elif global_manager.settings_builder.task == "summarization": - llm_tuner = Seq2SeqFinetuner( - model_name=global_manager.settings_builder.model_name, - compute_specs=global_manager.settings_builder.compute_profile, - pipeline_task="summarization" - ) - elif global_manager.settings_builder.task == "extractive-question-answering": - llm_tuner = QuestionAnsweringTuner( + + # Use provider adapter if a provider is explicitly specified + # Otherwise fall back to legacy tuners for backward compatibility + if provider and provider != "huggingface": + # Use provider adapter for non-HuggingFace providers + llm_tuner = ProviderFinetuner( model_name=global_manager.settings_builder.model_name, - compute_specs=global_manager.settings_builder.compute_profile, - pipeline_task="question-answering" + task=global_manager.settings_builder.task, + provider=provider, + compute_specs=global_manager.settings_builder.compute_profile ) else: - raise HTTPException( - status_code=400, - detail=f"Invalid task. Must be one of {VALID_TASKS_STR}." - ) + # Legacy path - use existing tuner classes for HuggingFace + # This maintains backward compatibility + if global_manager.settings_builder.task == "text-generation": + llm_tuner = CausalLLMFinetuner( + model_name=global_manager.settings_builder.model_name, + compute_specs=global_manager.settings_builder.compute_profile, + pipeline_task="text-generation" + ) + elif global_manager.settings_builder.task == "summarization": + llm_tuner = Seq2SeqFinetuner( + model_name=global_manager.settings_builder.model_name, + compute_specs=global_manager.settings_builder.compute_profile, + pipeline_task="summarization" + ) + elif global_manager.settings_builder.task == "extractive-question-answering": + llm_tuner = QuestionAnsweringTuner( + model_name=global_manager.settings_builder.model_name, + compute_specs=global_manager.settings_builder.compute_profile, + pipeline_task="question-answering" + ) + else: + raise HTTPException( + status_code=400, + detail=f"Invalid task. Must be one of {VALID_TASKS_STR}." + ) llm_tuner.set_settings(**global_manager.settings_builder.get_settings()) diff --git a/ModelForge/utilities/finetuning/provider_adapter.py b/ModelForge/utilities/finetuning/provider_adapter.py new file mode 100644 index 0000000..c2157c7 --- /dev/null +++ b/ModelForge/utilities/finetuning/provider_adapter.py @@ -0,0 +1,205 @@ +""" +Adapter for integrating provider-based finetuning with existing infrastructure. + +This module provides a bridge between the new provider system and the +existing Finetuner-based workflow to maintain backward compatibility. +""" + +import json +import os +from typing import Optional +from datasets import Dataset + +from .providers import get_provider, ProviderRegistry +from ...globals.globals_instance import global_manager + + +class ProviderFinetuner: + """ + Adapter class that uses the provider system while maintaining + compatibility with the existing finetuning workflow. + """ + + def __init__( + self, + model_name: str, + task: str, + provider: str = "huggingface", + compute_specs: str = "low_end" + ) -> None: + """ + Initialize provider-based finetuner. + + Args: + model_name: Model identifier + task: Task type (text-generation, summarization, extractive-question-answering) + provider: Provider name (huggingface, unsloth) + compute_specs: Hardware profile + """ + self.model_name = model_name + self.task = task + self.provider_name = provider + self.compute_specs = compute_specs + + # Get provider instance + self.provider = get_provider( + provider_name=provider, + model_name=model_name, + task=task, + compute_specs=compute_specs + ) + + # Settings that will be set by set_settings() + self.output_dir: Optional[str] = None + self.fine_tuned_name: Optional[str] = None + self.dataset_path: Optional[str] = None + + # Map task to pipeline task string + self.pipeline_task = self._map_task_to_pipeline(task) + + def _map_task_to_pipeline(self, task: str) -> str: + """Map task to pipeline task string.""" + task_map = { + "text-generation": "text-generation", + "summarization": "summarization", + "extractive-question-answering": "question-answering", + } + return task_map.get(task, task) + + def set_settings(self, **kwargs) -> None: + """ + Set training settings for the provider. + + Args: + **kwargs: Settings dictionary + """ + # Generate output paths using existing logic + from .Finetuner import Finetuner + + uid = Finetuner.gen_uuid() + safe_model_name = self.model_name.replace('/', '-').replace('\\', '-') + + # Use FileManager default directories + default_dirs = global_manager.file_manager.return_default_dirs() + self.fine_tuned_name = f"{default_dirs['models']}/{safe_model_name}_{uid}" + self.output_dir = f"{default_dirs['model_checkpoints']}/{safe_model_name}_{uid}" + + # Add output paths to settings + kwargs["output_dir"] = self.output_dir + kwargs["fine_tuned_name"] = self.fine_tuned_name + + # Set provider settings + self.provider.set_settings(**kwargs) + + def load_dataset(self, dataset_path: str) -> None: + """ + Load and prepare the dataset. + + Args: + dataset_path: Path to dataset file + """ + self.dataset_path = dataset_path + self.provider.prepare_dataset(dataset_path) + + def finetune(self) -> bool | str: + """ + Execute the fine-tuning process. + + Returns: + Path to saved model if successful, False otherwise + """ + try: + # Load model + self.provider.load_model(**self.provider.settings) + + # Ensure dataset is loaded + if self.provider.dataset is None: + if self.dataset_path: + self.load_dataset(self.dataset_path) + else: + raise ValueError("Dataset must be loaded before training") + + # Train + model_path = self.provider.train(**self.provider.settings) + + # Build config file for playground compatibility + config_file_result = self._build_config_file( + model_path, + self.pipeline_task + ) + + if not config_file_result: + print("Warning: Failed to create config file. Model may not work in playground.") + + # Report finish + self._report_finish() + + return model_path + + except Exception as e: + print(f"Fine-tuning failed: {e}") + self._report_finish(error=True, message=str(e)) + return False + + def _build_config_file(self, config_dir: str, pipeline_task: str) -> bool: + """ + Build configuration file for the fine-tuned model. + + Args: + config_dir: Directory to save config + pipeline_task: Pipeline task string + + Returns: + True if successful + """ + # Determine model class based on provider and task + model_class_map = { + "huggingface": { + "text-generation": "AutoPeftModelForCausalLM", + "summarization": "AutoPeftModelForSeq2SeqLM", + "question-answering": "AutoPeftModelForCausalLM", + }, + "unsloth": { + "text-generation": "AutoPeftModelForCausalLM", + "summarization": "AutoPeftModelForSeq2SeqLM", + "question-answering": "AutoPeftModelForCausalLM", + } + } + + model_class = model_class_map.get(self.provider_name, {}).get( + pipeline_task, + "AutoPeftModelForCausalLM" + ) + + try: + config_path = os.path.join(config_dir, "modelforge_config.json") + with open(config_path, "w") as f: + config = { + "model_class": model_class, + "pipeline_task": pipeline_task, + "provider": self.provider_name, + } + json.dump(config, f, indent=4) + print(f"Configuration file saved to {config_path}") + return True + except Exception as e: + print(f"Error saving configuration file: {e}") + return False + + def _report_finish(self, error: bool = False, message: Optional[str] = None) -> None: + """ + Report completion of fine-tuning. + + Args: + error: True if an error occurred + message: Error message if applicable + """ + print("*" * 100) + if not error: + print(f"Model fine-tuned successfully using {self.provider_name}!") + print(f"Model saved to {self.fine_tuned_name}") + print("Try out your new model in our chat playground!") + else: + print("Model fine-tuning failed!") + print(f"Error: {message}") + print("*" * 100) diff --git a/ModelForge/utilities/finetuning/providers/__init__.py b/ModelForge/utilities/finetuning/providers/__init__.py new file mode 100644 index 0000000..86496bb --- /dev/null +++ b/ModelForge/utilities/finetuning/providers/__init__.py @@ -0,0 +1,23 @@ +""" +Provider abstraction layer for fine-tuning backends. + +This module provides a unified interface for different fine-tuning providers +(e.g., HuggingFace, Unsloth) to enable pluggable backend implementations. +""" + +from .base_provider import FinetuningProvider +from .huggingface_provider import HuggingFaceProvider +from .unsloth_provider import UnslothProvider +from .provider_registry import ProviderRegistry, get_provider + +# Register providers +ProviderRegistry.register(HuggingFaceProvider) +ProviderRegistry.register(UnslothProvider) + +__all__ = [ + "FinetuningProvider", + "HuggingFaceProvider", + "UnslothProvider", + "ProviderRegistry", + "get_provider", +] diff --git a/ModelForge/utilities/finetuning/providers/base_provider.py b/ModelForge/utilities/finetuning/providers/base_provider.py new file mode 100644 index 0000000..ec9975a --- /dev/null +++ b/ModelForge/utilities/finetuning/providers/base_provider.py @@ -0,0 +1,173 @@ +""" +Base provider interface for fine-tuning implementations. + +This module defines the abstract interface that all fine-tuning providers +must implement to ensure consistent behavior across different backends. +""" + +from abc import ABC, abstractmethod +from typing import Dict, List, Optional, Any, Tuple +from datasets import Dataset + + +class FinetuningProvider(ABC): + """ + Abstract base class defining the interface for fine-tuning providers. + + All fine-tuning providers (HuggingFace, Unsloth, etc.) must implement + this interface to ensure compatibility with the ModelForge pipeline. + """ + + def __init__( + self, + model_name: str, + task: str, + compute_specs: str = "low_end" + ) -> None: + """ + Initialize the fine-tuning provider. + + Args: + model_name: Name or path of the model to fine-tune + task: Task type (text-generation, summarization, extractive-question-answering) + compute_specs: Hardware profile (low_end, mid_range, high_end) + """ + self.model_name = model_name + self.task = task + self.compute_specs = compute_specs + self.model = None + self.tokenizer = None + self.dataset: Optional[Dataset] = None + self.settings: Dict[str, Any] = {} + + @abstractmethod + def load_model(self, **kwargs) -> Tuple[Any, Any]: + """ + Load the model and tokenizer with provider-specific configurations. + + Args: + **kwargs: Provider-specific model loading parameters + + Returns: + Tuple of (model, tokenizer) + + Raises: + Exception: If model loading fails + """ + pass + + @abstractmethod + def prepare_dataset(self, dataset_path: str, **kwargs) -> Dataset: + """ + Load and prepare the dataset for training. + + Args: + dataset_path: Path to the dataset file + **kwargs: Provider-specific dataset preparation parameters + + Returns: + Prepared dataset ready for training + + Raises: + Exception: If dataset loading or preparation fails + """ + pass + + @abstractmethod + def train(self, **kwargs) -> str: + """ + Execute the fine-tuning process. + + Args: + **kwargs: Provider-specific training parameters + + Returns: + Path to the saved fine-tuned model + + Raises: + Exception: If training fails + """ + pass + + @abstractmethod + def export_model(self, output_path: str, **kwargs) -> bool: + """ + Export the fine-tuned model to the specified path. + + Args: + output_path: Directory to save the exported model + **kwargs: Provider-specific export parameters + + Returns: + True if export succeeds, False otherwise + """ + pass + + @abstractmethod + def get_supported_hyperparameters(self) -> List[str]: + """ + Return a list of hyperparameters supported by this provider. + + Returns: + List of hyperparameter names + """ + pass + + @abstractmethod + def validate_settings(self, settings: Dict[str, Any]) -> Dict[str, Any]: + """ + Validate and sanitize provider-specific settings. + + Args: + settings: Dictionary of hyperparameters and configurations + + Returns: + Validated settings dictionary + + Raises: + ValueError: If settings are invalid + """ + pass + + @classmethod + @abstractmethod + def get_provider_name(cls) -> str: + """ + Return the canonical name of this provider. + + Returns: + Provider name (e.g., "huggingface", "unsloth") + """ + pass + + @classmethod + @abstractmethod + def get_provider_description(cls) -> str: + """ + Return a human-readable description of this provider. + + Returns: + Provider description + """ + pass + + @classmethod + @abstractmethod + def is_available(cls) -> bool: + """ + Check if this provider's dependencies are installed and available. + + Returns: + True if provider can be used, False otherwise + """ + pass + + def set_settings(self, **kwargs) -> None: + """ + Set training settings from keyword arguments. + + Args: + **kwargs: Settings to apply + """ + validated_settings = self.validate_settings(kwargs) + self.settings.update(validated_settings) diff --git a/ModelForge/utilities/finetuning/providers/huggingface_provider.py b/ModelForge/utilities/finetuning/providers/huggingface_provider.py new file mode 100644 index 0000000..e19a0f9 --- /dev/null +++ b/ModelForge/utilities/finetuning/providers/huggingface_provider.py @@ -0,0 +1,386 @@ +""" +HuggingFace provider implementation for fine-tuning. + +This module provides HuggingFace Transformers-based fine-tuning, +maintaining backward compatibility with existing ModelForge workflows. +""" + +import os +from typing import Dict, List, Any, Tuple, Optional +import torch +from datasets import Dataset, load_dataset +from transformers import ( + AutoModelForCausalLM, + AutoModelForSeq2SeqLM, + AutoTokenizer, + BitsAndBytesConfig, + TrainerCallback, +) +from peft import LoraConfig, TaskType, get_peft_model +from trl import SFTTrainer, SFTConfig + +from .base_provider import FinetuningProvider + + +class HuggingFaceProgressCallback(TrainerCallback): + """ + Callback to update global finetuning status during HuggingFace training. + """ + + def __init__(self): + super().__init__() + from ....globals.globals_instance import global_manager + self.global_manager = global_manager + + def on_log(self, args, state, control, logs=None, **kwargs): + """Called when logging happens during training.""" + if state.max_steps <= 0: + return + + progress = min(95, int((state.global_step / state.max_steps) * 100)) + self.global_manager.finetuning_status["progress"] = progress + self.global_manager.finetuning_status["message"] = ( + f"Training step {state.global_step}/{state.max_steps}" + ) + + def on_train_end(self, args, state, control, **kwargs): + """Called at the end of training.""" + self.global_manager.finetuning_status["progress"] = 100 + self.global_manager.finetuning_status["message"] = "Training completed!" + + +class HuggingFaceProvider(FinetuningProvider): + """ + HuggingFace Transformers provider for fine-tuning. + + Implements the FinetuningProvider interface using HuggingFace's + transformers, peft, and trl libraries. + """ + + # Task type mappings for PEFT + TASK_TYPE_MAP = { + "text-generation": TaskType.CAUSAL_LM, + "summarization": TaskType.SEQ_2_SEQ_LM, + "extractive-question-answering": TaskType.QUESTION_ANS, + } + + # Model class mappings + MODEL_CLASS_MAP = { + "text-generation": (AutoModelForCausalLM, "AutoPeftModelForCausalLM"), + "summarization": (AutoModelForSeq2SeqLM, "AutoPeftModelForSeq2SeqLM"), + "extractive-question-answering": (AutoModelForCausalLM, "AutoPeftModelForCausalLM"), + } + + def __init__( + self, + model_name: str, + task: str, + compute_specs: str = "low_end" + ) -> None: + """ + Initialize HuggingFace provider. + + Args: + model_name: HuggingFace model identifier + task: Fine-tuning task type + compute_specs: Hardware profile + """ + super().__init__(model_name, task, compute_specs) + self.peft_task_type = self.TASK_TYPE_MAP.get(task) + self.model_class, self.peft_class_name = self.MODEL_CLASS_MAP.get(task, (None, None)) + self.output_dir: Optional[str] = None + self.fine_tuned_name: Optional[str] = None + + def load_model(self, **kwargs) -> Tuple[Any, Any]: + """ + Load HuggingFace model and tokenizer with quantization. + + Args: + **kwargs: Settings including quantization config + + Returns: + Tuple of (model, tokenizer) + """ + if self.model_class is None: + raise ValueError(f"Unsupported task type: {self.task}") + + # Prepare quantization config + bits_n_bytes_config = None + use_4bit = kwargs.get("use_4bit", False) + use_8bit = kwargs.get("use_8bit", False) + + if use_4bit: + compute_dtype = getattr(torch, kwargs.get("bnb_4bit_compute_dtype", "float16")) + bits_n_bytes_config = BitsAndBytesConfig( + load_in_4bit=True, + bnb_4bit_quant_type=kwargs.get("bnb_4bit_quant_type", "nf4"), + bnb_4bit_compute_dtype=compute_dtype, + bnb_4bit_use_double_quant=kwargs.get("use_nested_quant", False), + ) + elif use_8bit: + bits_n_bytes_config = BitsAndBytesConfig( + load_in_8bit=True, + ) + + # Load model + device_map = kwargs.get("device_map", {"": 0}) + + if bits_n_bytes_config: + model = self.model_class.from_pretrained( + self.model_name, + quantization_config=bits_n_bytes_config, + device_map=device_map, + use_cache=False, + ) + else: + model = self.model_class.from_pretrained( + self.model_name, + device_map=device_map, + use_cache=False, + ) + + # Load tokenizer + tokenizer = AutoTokenizer.from_pretrained( + self.model_name, + trust_remote_code=True + ) + tokenizer.pad_token = tokenizer.eos_token + tokenizer.padding_side = "right" + + self.model = model + self.tokenizer = tokenizer + + return model, tokenizer + + def prepare_dataset(self, dataset_path: str, **kwargs) -> Dataset: + """ + Load and format dataset for HuggingFace training. + + Args: + dataset_path: Path to dataset file + **kwargs: Additional dataset preparation parameters + + Returns: + Formatted dataset + """ + dataset = load_dataset("json", data_files=dataset_path, split="train") + + # Format based on task type + if self.task == "text-generation": + dataset = dataset.rename_column("input", "prompt") + dataset = dataset.rename_column("output", "completion") + dataset = dataset.map(self._format_text_generation_example) + elif self.task == "summarization": + keys = dataset.column_names + dataset = dataset.map(lambda x: self._format_summarization_example(x, keys)) + dataset = dataset.remove_columns(keys) + elif self.task == "extractive-question-answering": + keys = dataset.column_names + dataset = dataset.map(lambda x: self._format_qa_example(x, keys)) + dataset = dataset.remove_columns(keys) + + self.dataset = dataset + return dataset + + def _format_text_generation_example(self, example: dict) -> Dict[str, str]: + """Format example for text generation.""" + return { + "prompt": "USER:" + example.get("prompt", ""), + "completion": "ASSISTANT: " + example.get("completion", "") + "<|endoftext|>" + } + + def _format_summarization_example(self, example: dict, keys: List[str]) -> Dict[str, str]: + """Format example for summarization.""" + if len(keys) < 2: + keys = ["article", "summary"] + return { + "text": f''' + ["role": "system", "content": "You are a text summarization assistant."], + ["role": "user", "content": {example[keys[0]]}], + ["role": "assistant", "content": {example[keys[1]]}] + ''' + } + + def _format_qa_example(self, example: dict, keys: List[str]) -> Dict[str, str]: + """Format example for question answering.""" + if len(keys) < 3: + keys = ["context", "question", "answer"] + return { + "text": f''' + ["role": "system", "content": "You are a question answering assistant."], + ["role": "user", "content": "Context: {example[keys[0]]}\nQuestion: {example[keys[1]]}"], + ["role": "assistant", "content": {example[keys[2]]}] + ''' + } + + def train(self, **kwargs) -> str: + """ + Execute HuggingFace fine-tuning with PEFT/LoRA. + + Args: + **kwargs: Training configuration + + Returns: + Path to saved model + """ + # Ensure model and dataset are loaded + if self.model is None or self.tokenizer is None: + self.load_model(**kwargs) + + if self.dataset is None: + raise ValueError("Dataset must be prepared before training") + + # Configure LoRA + lora_config = LoraConfig( + lora_alpha=kwargs.get("lora_alpha", 32), + lora_dropout=kwargs.get("lora_dropout", 0.1), + r=kwargs.get("lora_r", 16), + bias="none", + task_type=self.peft_task_type, + target_modules='all-linear', + ) + + # Apply PEFT + model = get_peft_model(self.model, lora_config) + + # Configure training + training_args = SFTConfig( + output_dir=self.output_dir or "./model_checkpoints", + num_train_epochs=kwargs.get("num_train_epochs", 1), + per_device_train_batch_size=kwargs.get("per_device_train_batch_size", 1), + gradient_accumulation_steps=kwargs.get("gradient_accumulation_steps", 4), + optim=kwargs.get("optim", "paged_adamw_32bit"), + save_steps=kwargs.get("save_steps", 0), + logging_steps=kwargs.get("logging_steps", 25), + learning_rate=kwargs.get("learning_rate", 2e-4), + warmup_ratio=kwargs.get("warmup_ratio", 0.03), + weight_decay=kwargs.get("weight_decay", 0.001), + fp16=kwargs.get("fp16", False), + bf16=kwargs.get("bf16", False), + max_grad_norm=kwargs.get("max_grad_norm", 0.3), + max_steps=kwargs.get("max_steps", -1), + group_by_length=kwargs.get("group_by_length", True), + lr_scheduler_type=kwargs.get("lr_scheduler_type", "cosine"), + report_to="tensorboard", + logging_dir="./training_logs", + max_length=None, + ) + + # Create trainer + trainer = SFTTrainer( + model=model, + train_dataset=self.dataset, + args=training_args, + callbacks=[HuggingFaceProgressCallback()], + ) + + # Train + trainer.train() + + # Save model + save_path = self.fine_tuned_name or self.output_dir + trainer.model.save_pretrained(save_path) + + return save_path + + def export_model(self, output_path: str, **kwargs) -> bool: + """ + Export the fine-tuned model (already saved during training). + + Args: + output_path: Path to export the model + **kwargs: Additional export parameters + + Returns: + True if successful + """ + # HuggingFace models are already saved during training + # This method exists for interface compatibility + return True + + def get_supported_hyperparameters(self) -> List[str]: + """ + Return list of supported hyperparameters. + + Returns: + List of hyperparameter names + """ + return [ + "num_train_epochs", + "lora_r", + "lora_alpha", + "lora_dropout", + "use_4bit", + "use_8bit", + "bnb_4bit_compute_dtype", + "bnb_4bit_quant_type", + "use_nested_quant", + "fp16", + "bf16", + "per_device_train_batch_size", + "per_device_eval_batch_size", + "gradient_accumulation_steps", + "gradient_checkpointing", + "max_grad_norm", + "learning_rate", + "weight_decay", + "optim", + "lr_scheduler_type", + "max_steps", + "warmup_ratio", + "group_by_length", + "packing", + "device_map", + "max_seq_length", + ] + + def validate_settings(self, settings: Dict[str, Any]) -> Dict[str, Any]: + """ + Validate HuggingFace-specific settings. + + Args: + settings: Settings dictionary + + Returns: + Validated settings + """ + # Basic validation - detailed validation happens in the router + validated = {} + + for key, value in settings.items(): + if key in self.get_supported_hyperparameters(): + validated[key] = value + + # Extract output paths if present + if "output_dir" in settings: + self.output_dir = settings["output_dir"] + if "fine_tuned_name" in settings: + self.fine_tuned_name = settings["fine_tuned_name"] + + return validated + + @classmethod + def get_provider_name(cls) -> str: + """Return provider name.""" + return "huggingface" + + @classmethod + def get_provider_description(cls) -> str: + """Return provider description.""" + return "HuggingFace Transformers with PEFT/LoRA fine-tuning" + + @classmethod + def is_available(cls) -> bool: + """ + Check if HuggingFace dependencies are available. + + Returns: + True if available + """ + try: + import transformers + import peft + import trl + return True + except ImportError: + return False diff --git a/ModelForge/utilities/finetuning/providers/provider_registry.py b/ModelForge/utilities/finetuning/providers/provider_registry.py new file mode 100644 index 0000000..ec56333 --- /dev/null +++ b/ModelForge/utilities/finetuning/providers/provider_registry.py @@ -0,0 +1,142 @@ +""" +Provider registry for managing and selecting fine-tuning providers. + +This module provides a centralized registry for all available fine-tuning +providers, enabling runtime provider selection and discovery. +""" + +from typing import Dict, Type, List, Optional +from .base_provider import FinetuningProvider + + +class ProviderRegistry: + """ + Central registry for managing fine-tuning provider implementations. + """ + + _providers: Dict[str, Type[FinetuningProvider]] = {} + + @classmethod + def register(cls, provider_class: Type[FinetuningProvider]) -> None: + """ + Register a fine-tuning provider. + + Args: + provider_class: Provider class to register + + Raises: + ValueError: If provider is already registered + """ + provider_name = provider_class.get_provider_name() + + if provider_name in cls._providers: + raise ValueError(f"Provider '{provider_name}' is already registered") + + cls._providers[provider_name] = provider_class + + @classmethod + def get(cls, provider_name: str) -> Optional[Type[FinetuningProvider]]: + """ + Get a provider class by name. + + Args: + provider_name: Name of the provider + + Returns: + Provider class or None if not found + """ + return cls._providers.get(provider_name) + + @classmethod + def list_available(cls) -> List[Dict[str, str]]: + """ + List all available (installed) providers. + + Returns: + List of dictionaries containing provider information + """ + available = [] + for name, provider_class in cls._providers.items(): + if provider_class.is_available(): + available.append({ + "name": name, + "description": provider_class.get_provider_description(), + }) + return available + + @classmethod + def list_all(cls) -> List[Dict[str, str]]: + """ + List all registered providers (including unavailable ones). + + Returns: + List of dictionaries containing provider information + """ + all_providers = [] + for name, provider_class in cls._providers.items(): + all_providers.append({ + "name": name, + "description": provider_class.get_provider_description(), + "available": provider_class.is_available(), + }) + return all_providers + + @classmethod + def is_available(cls, provider_name: str) -> bool: + """ + Check if a provider is available for use. + + Args: + provider_name: Name of the provider + + Returns: + True if provider exists and is available, False otherwise + """ + provider_class = cls.get(provider_name) + if provider_class is None: + return False + return provider_class.is_available() + + +def get_provider( + provider_name: str, + model_name: str, + task: str, + compute_specs: str = "low_end" +) -> FinetuningProvider: + """ + Factory function to instantiate a provider. + + Args: + provider_name: Name of the provider to instantiate + model_name: Model name or path + task: Fine-tuning task + compute_specs: Hardware profile + + Returns: + Instantiated provider + + Raises: + ValueError: If provider not found or not available + """ + provider_class = ProviderRegistry.get(provider_name) + + if provider_class is None: + available = ProviderRegistry.list_available() + available_names = [p["name"] for p in available] + raise ValueError( + f"Provider '{provider_name}' not found. " + f"Available providers: {', '.join(available_names)}" + ) + + if not provider_class.is_available(): + raise ValueError( + f"Provider '{provider_name}' is registered but not available. " + f"Please ensure all required dependencies are installed." + ) + + return provider_class( + model_name=model_name, + task=task, + compute_specs=compute_specs + ) diff --git a/ModelForge/utilities/finetuning/providers/unsloth_provider.py b/ModelForge/utilities/finetuning/providers/unsloth_provider.py new file mode 100644 index 0000000..2758e72 --- /dev/null +++ b/ModelForge/utilities/finetuning/providers/unsloth_provider.py @@ -0,0 +1,380 @@ +""" +Unsloth AI provider implementation for fine-tuning. + +This module provides Unsloth-based fine-tuning with optimized +memory usage and faster training speeds compared to standard HuggingFace. +""" + +import os +from typing import Dict, List, Any, Tuple, Optional +from datasets import Dataset, load_dataset + +from .base_provider import FinetuningProvider + + +class UnslothProvider(FinetuningProvider): + """ + Unsloth AI provider for optimized fine-tuning. + + Implements the FinetuningProvider interface using Unsloth's + optimized training infrastructure for faster and more memory-efficient + fine-tuning of large language models. + """ + + # Task type mappings for Unsloth + TASK_TYPE_MAP = { + "text-generation": "causal", + "summarization": "seq2seq", + "extractive-question-answering": "causal", + } + + def __init__( + self, + model_name: str, + task: str, + compute_specs: str = "low_end" + ) -> None: + """ + Initialize Unsloth provider. + + Args: + model_name: Model identifier (HuggingFace format) + task: Fine-tuning task type + compute_specs: Hardware profile + """ + super().__init__(model_name, task, compute_specs) + self.unsloth_task = self.TASK_TYPE_MAP.get(task) + self.output_dir: Optional[str] = None + self.fine_tuned_name: Optional[str] = None + + def load_model(self, **kwargs) -> Tuple[Any, Any]: + """ + Load model using Unsloth's FastLanguageModel. + + Args: + **kwargs: Settings including max_seq_length, quantization, etc. + + Returns: + Tuple of (model, tokenizer) + """ + try: + from unsloth import FastLanguageModel + except ImportError: + raise ImportError( + "Unsloth is not installed. Install it with: " + "pip install unsloth" + ) + + # Determine quantization settings + load_in_4bit = kwargs.get("use_4bit", False) or kwargs.get("load_in_4bit", False) + load_in_8bit = kwargs.get("use_8bit", False) or kwargs.get("load_in_8bit", False) + + # Unsloth-specific parameters + max_seq_length = kwargs.get("max_seq_length", 2048) + if max_seq_length == -1 or max_seq_length is None: + max_seq_length = 2048 # Unsloth default + + dtype = None # Auto-detect + if kwargs.get("bf16", False): + import torch + dtype = torch.bfloat16 + elif kwargs.get("fp16", False): + import torch + dtype = torch.float16 + + # Load model with Unsloth optimizations + model, tokenizer = FastLanguageModel.from_pretrained( + model_name=self.model_name, + max_seq_length=max_seq_length, + dtype=dtype, + load_in_4bit=load_in_4bit, + ) + + self.model = model + self.tokenizer = tokenizer + + return model, tokenizer + + def prepare_dataset(self, dataset_path: str, **kwargs) -> Dataset: + """ + Load and format dataset for Unsloth training. + + Args: + dataset_path: Path to dataset file + **kwargs: Additional dataset preparation parameters + + Returns: + Formatted dataset + """ + dataset = load_dataset("json", data_files=dataset_path, split="train") + + # Format based on task type + if self.task == "text-generation": + dataset = dataset.rename_column("input", "prompt") + dataset = dataset.rename_column("output", "completion") + dataset = dataset.map(self._format_text_generation_example) + elif self.task == "summarization": + keys = dataset.column_names + dataset = dataset.map(lambda x: self._format_summarization_example(x, keys)) + dataset = dataset.remove_columns(keys) + elif self.task == "extractive-question-answering": + keys = dataset.column_names + dataset = dataset.map(lambda x: self._format_qa_example(x, keys)) + dataset = dataset.remove_columns(keys) + + self.dataset = dataset + return dataset + + def _format_text_generation_example(self, example: dict) -> Dict[str, str]: + """Format example for text generation with Unsloth.""" + # Unsloth uses a more standard chat format + return { + "text": f"### User:\n{example.get('prompt', '')}\n\n### Assistant:\n{example.get('completion', '')}<|endoftext|>" + } + + def _format_summarization_example(self, example: dict, keys: List[str]) -> Dict[str, str]: + """Format example for summarization with Unsloth.""" + if len(keys) < 2: + keys = ["article", "summary"] + return { + "text": f"### Article:\n{example[keys[0]]}\n\n### Summary:\n{example[keys[1]]}<|endoftext|>" + } + + def _format_qa_example(self, example: dict, keys: List[str]) -> Dict[str, str]: + """Format example for question answering with Unsloth.""" + if len(keys) < 3: + keys = ["context", "question", "answer"] + return { + "text": f"### Context:\n{example[keys[0]]}\n\n### Question:\n{example[keys[1]]}\n\n### Answer:\n{example[keys[2]]}<|endoftext|>" + } + + def train(self, **kwargs) -> str: + """ + Execute Unsloth fine-tuning with optimized LoRA/QLoRA. + + Args: + **kwargs: Training configuration + + Returns: + Path to saved model + """ + try: + from unsloth import FastLanguageModel + from trl import SFTTrainer + from transformers import TrainingArguments, TrainerCallback + except ImportError as e: + raise ImportError(f"Required package not available: {e}") + + # Ensure model and dataset are loaded + if self.model is None or self.tokenizer is None: + self.load_model(**kwargs) + + if self.dataset is None: + raise ValueError("Dataset must be prepared before training") + + # Apply Unsloth PEFT (optimized LoRA) + model = FastLanguageModel.get_peft_model( + self.model, + r=kwargs.get("lora_r", 16), + target_modules=["q_proj", "k_proj", "v_proj", "o_proj", + "gate_proj", "up_proj", "down_proj"], + lora_alpha=kwargs.get("lora_alpha", 32), + lora_dropout=kwargs.get("lora_dropout", 0.1), + bias="none", + use_gradient_checkpointing=kwargs.get("gradient_checkpointing", True), + random_state=3407, + use_rslora=False, # Rank stabilized LoRA + loftq_config=None, # LoftQ quantization + ) + + # Progress callback + class UnslothProgressCallback(TrainerCallback): + """Callback to update global finetuning status during Unsloth training.""" + + def __init__(self): + super().__init__() + from ....globals.globals_instance import global_manager + self.global_manager = global_manager + + def on_log(self, args, state, control, logs=None, **kwargs): + """Called when logging happens during training.""" + if state.max_steps <= 0: + return + + progress = min(95, int((state.global_step / state.max_steps) * 100)) + self.global_manager.finetuning_status["progress"] = progress + self.global_manager.finetuning_status["message"] = ( + f"Training step {state.global_step}/{state.max_steps} (Unsloth)" + ) + + def on_train_end(self, args, state, control, **kwargs): + """Called at the end of training.""" + self.global_manager.finetuning_status["progress"] = 100 + self.global_manager.finetuning_status["message"] = "Training completed with Unsloth!" + + # Configure training arguments + training_args = TrainingArguments( + output_dir=self.output_dir or "./model_checkpoints", + num_train_epochs=kwargs.get("num_train_epochs", 1), + per_device_train_batch_size=kwargs.get("per_device_train_batch_size", 2), + gradient_accumulation_steps=kwargs.get("gradient_accumulation_steps", 4), + warmup_steps=int(kwargs.get("warmup_ratio", 0.03) * 100), # Approximate + learning_rate=kwargs.get("learning_rate", 2e-4), + fp16=kwargs.get("fp16", False), + bf16=kwargs.get("bf16", False), + logging_steps=kwargs.get("logging_steps", 1), + optim=kwargs.get("optim", "adamw_8bit"), + weight_decay=kwargs.get("weight_decay", 0.01), + lr_scheduler_type=kwargs.get("lr_scheduler_type", "linear"), + seed=3407, + save_steps=kwargs.get("save_steps", 0), + max_steps=kwargs.get("max_steps", -1), + report_to="tensorboard", + logging_dir="./training_logs", + ) + + # Create SFT trainer with Unsloth optimizations + trainer = SFTTrainer( + model=model, + tokenizer=self.tokenizer, + train_dataset=self.dataset, + dataset_text_field="text", + max_seq_length=kwargs.get("max_seq_length", 2048) if kwargs.get("max_seq_length", 2048) != -1 else 2048, + dataset_num_proc=2, + packing=kwargs.get("packing", False), + args=training_args, + callbacks=[UnslothProgressCallback()], + ) + + # Train with Unsloth optimizations + trainer.train() + + # Save model + save_path = self.fine_tuned_name or self.output_dir + model.save_pretrained(save_path) + self.tokenizer.save_pretrained(save_path) + + return save_path + + def export_model(self, output_path: str, **kwargs) -> bool: + """ + Export the fine-tuned Unsloth model. + + Supports multiple export formats including HuggingFace format. + Note: GGUF export requires additional Unsloth methods that may not be + available in all versions. Currently defaults to HuggingFace format. + + Args: + output_path: Path to export the model + **kwargs: Export parameters (format, quantization_method, etc.) + + Returns: + True if successful + """ + export_format = kwargs.get("export_format", "huggingface") + + if export_format == "huggingface": + # Already saved in HuggingFace format during training + return True + else: + # Other export formats (e.g., GGUF) may be supported in future Unsloth versions + # For now, return False for unsupported formats + print(f"Warning: Export format '{export_format}' is not yet supported. Model saved in HuggingFace format.") + return False + + def get_supported_hyperparameters(self) -> List[str]: + """ + Return list of Unsloth-supported hyperparameters. + + Returns: + List of hyperparameter names + """ + return [ + # Standard hyperparameters + "num_train_epochs", + "lora_r", + "lora_alpha", + "lora_dropout", + "use_4bit", + "load_in_4bit", + "fp16", + "bf16", + "per_device_train_batch_size", + "gradient_accumulation_steps", + "gradient_checkpointing", + "learning_rate", + "weight_decay", + "optim", + "lr_scheduler_type", + "max_steps", + "warmup_ratio", + "packing", + "max_seq_length", + # Unsloth-specific + "use_rslora", + "use_gradient_checkpointing", + ] + + def validate_settings(self, settings: Dict[str, Any]) -> Dict[str, Any]: + """ + Validate Unsloth-specific settings. + + Args: + settings: Settings dictionary + + Returns: + Validated settings + """ + validated = {} + + for key, value in settings.items(): + if key in self.get_supported_hyperparameters(): + validated[key] = value + + # Extract output paths if present + if "output_dir" in settings: + self.output_dir = settings["output_dir"] + if "fine_tuned_name" in settings: + self.fine_tuned_name = settings["fine_tuned_name"] + + # Unsloth-specific validations + if "max_seq_length" in validated: + if validated["max_seq_length"] == -1 or validated["max_seq_length"] is None: + validated["max_seq_length"] = 2048 + + # Unsloth works best with specific optimizers + if "optim" in validated: + # Map to Unsloth-compatible optimizers + optimizer_map = { + "paged_adamw_32bit": "adamw_8bit", + "paged_adamw_8bit": "adamw_8bit", + "adamw_torch": "adamw_torch", + "adamw_hf": "adamw_torch", + } + validated["optim"] = optimizer_map.get(validated["optim"], "adamw_8bit") + + return validated + + @classmethod + def get_provider_name(cls) -> str: + """Return provider name.""" + return "unsloth" + + @classmethod + def get_provider_description(cls) -> str: + """Return provider description.""" + return "Unsloth AI - 2x faster fine-tuning with reduced memory usage" + + @classmethod + def is_available(cls) -> bool: + """ + Check if Unsloth dependencies are available. + + Returns: + True if Unsloth is installed + """ + try: + import unsloth + return True + except ImportError: + return False diff --git a/ModelForge/utilities/finetuning/settings_builder.py b/ModelForge/utilities/finetuning/settings_builder.py index 9eb6e87..85eac3d 100644 --- a/ModelForge/utilities/finetuning/settings_builder.py +++ b/ModelForge/utilities/finetuning/settings_builder.py @@ -12,6 +12,7 @@ def __init__(self, task, model_name, compute_profile) -> None: self.dataset = None self.compute_profile = compute_profile self.is_custom_model = False + self.provider = "huggingface" # Default provider for backward compatibility self.lora_r = 16 self.lora_alpha = 32 self.lora_dropout = 0.1 @@ -56,6 +57,8 @@ def set_settings(self, settings_dict) -> None: for key, value in settings_dict.items(): if key == "dataset": self.dataset = value + elif key == "provider": + self.provider = value elif key == "max_seq_length": if value == -1: self.max_seq_length = None @@ -99,6 +102,7 @@ def get_settings(self) -> Dict[str, Union[str, float]]: return { "task": self.task, "model_name": self.model_name, + "provider": self.provider, "num_train_epochs": self.num_train_epochs, "compute_specs": self.compute_profile, "lora_r": self.lora_r, diff --git a/README.md b/README.md index 9b95a96..6fbe4ba 100644 --- a/README.md +++ b/README.md @@ -7,7 +7,8 @@ ## 🚀 **Features** - **GPU-Powered Finetuning**: Optimized for NVIDIA GPUs (even 4GB VRAM). -- **One-Click Workflow**: Upload data → Pick task → Train → Test. +- **Multiple Providers**: Choose between HuggingFace (standard) or Unsloth AI (2x faster, reduced memory) for fine-tuning. +- **One-Click Workflow**: Upload data → Pick task → Select provider → Train → Test. - **Hardware-Aware**: Auto-detects your GPU/CPU and recommends models. - **React UI**: No CLI or notebooks—just a friendly interface. @@ -79,6 +80,52 @@ To stop the application and free up resources, press `Ctrl+C` in the terminal ru {"input": "Enter the poem topic here...", "output": "Roses are red..."} ``` +## 🔧 **Fine-tuning Providers** + +ModelForge supports multiple fine-tuning providers, allowing you to choose the best backend for your needs: + +### **HuggingFace (Default)** +- **Description**: Standard fine-tuning using HuggingFace Transformers with PEFT/LoRA +- **Installation**: Included by default +- **Best for**: General use, maximum compatibility +- **Features**: 4-bit/8-bit quantization, gradient checkpointing, standard LoRA + +### **Unsloth AI (Optional)** +- **Description**: Optimized fine-tuning with 2x faster training and reduced memory usage +- **Installation**: + ```bash + pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git" + ``` +- **Best for**: Faster training, lower memory usage, larger models on limited hardware +- **Features**: Optimized LoRA/QLoRA, memory-efficient attention, faster gradient computation +- **Documentation**: [Unsloth Docs](https://docs.unsloth.ai/) + +### **Selecting a Provider** + +The provider can be selected when configuring your fine-tuning job: + +1. **Via UI**: Choose your provider from the dropdown in the settings page +2. **Via API**: Include the `provider` field in your settings: + ```json + { + "provider": "unsloth", + "task": "text-generation", + "model_name": "meta-llama/Llama-2-7b-hf", + ... + } + ``` + +**Note**: If a provider is not installed, ModelForge will fall back to HuggingFace automatically. + +### **Performance Comparison** + +| Provider | Training Speed | Memory Usage | Compatibility | +|------------|---------------|--------------|---------------| +| HuggingFace | 1x (baseline) | 1x (baseline)| Excellent | +| Unsloth | ~2x faster | ~30% less | Good | + +*Performance metrics may vary based on model size, hardware, and configuration.* + ## 🤝 **Contributing Model Recommendations** ModelForge uses a modular configuration system for model recommendations. Contributors can easily add new recommended models by adding configuration files to the `model_configs/` directory. Each hardware profile (low_end, mid_range, high_end) has its own configuration file where you can specify primary and alternative models for different tasks. diff --git a/pyproject.toml b/pyproject.toml index 418baaf..6ae60af 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -29,6 +29,11 @@ dependencies = [ "python-multipart", ] +[project.optional-dependencies] +unsloth = [ + "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git", +] + [tool.setuptools.packages.find] where = ["."] include = ["ModelForge*"]