├── LICENSE ├── README.md ├── models └── place-models-here.txt ├── modules ├── shared.py └── ui.py ├── repositories └── clone-audiocraft-here.txt ├── requirements.txt ├── results └── results-get-saved-here.txt ├── training └── datasets │ └── place-datasets-here.md ├── webui.py └── webuibatch.py /LICENSE: -------------------------------------------------------------------------------- 1 | GNU AFFERO GENERAL PUBLIC LICENSE 2 | Version 3, 19 November 2007 3 | 4 | Copyright (C) 2007 Free Software Foundation, Inc. 5 | Everyone is permitted to copy and distribute verbatim copies 6 | of this license document, but changing it is not allowed. 7 | 8 | Preamble 9 | 10 | The GNU Affero General Public License is a free, copyleft license for 11 | software and other kinds of works, specifically designed to ensure 12 | cooperation with the community in the case of network server software. 13 | 14 | The licenses for most software and other practical works are designed 15 | to take away your freedom to share and change the works. By contrast, 16 | our General Public Licenses are intended to guarantee your freedom to 17 | share and change all versions of a program--to make sure it remains free 18 | software for all its users. 19 | 20 | When we speak of free software, we are referring to freedom, not 21 | price. Our General Public Licenses are designed to make sure that you 22 | have the freedom to distribute copies of free software (and charge for 23 | them if you wish), that you receive source code or can get it if you 24 | want it, that you can change the software or use pieces of it in new 25 | free programs, and that you know you can do these things. 26 | 27 | Developers that use our General Public Licenses protect your rights 28 | with two steps: (1) assert copyright on the software, and (2) offer 29 | you this License which gives you legal permission to copy, distribute 30 | and/or modify the software. 31 | 32 | A secondary benefit of defending all users' freedom is that 33 | improvements made in alternate versions of the program, if they 34 | receive widespread use, become available for other developers to 35 | incorporate. Many developers of free software are heartened and 36 | encouraged by the resulting cooperation. However, in the case of 37 | software used on network servers, this result may fail to come about. 38 | The GNU General Public License permits making a modified version and 39 | letting the public access it on a server without ever releasing its 40 | source code to the public. 41 | 42 | The GNU Affero General Public License is designed specifically to 43 | ensure that, in such cases, the modified source code becomes available 44 | to the community. It requires the operator of a network server to 45 | provide the source code of the modified version running there to the 46 | users of that server. Therefore, public use of a modified version, on 47 | a publicly accessible server, gives the public access to the source 48 | code of the modified version. 49 | 50 | An older license, called the Affero General Public License and 51 | published by Affero, was designed to accomplish similar goals. This is 52 | a different license, not a version of the Affero GPL, but Affero has 53 | released a new version of the Affero GPL which permits relicensing under 54 | this license. 55 | 56 | The precise terms and conditions for copying, distribution and 57 | modification follow. 58 | 59 | TERMS AND CONDITIONS 60 | 61 | 0. Definitions. 62 | 63 | "This License" refers to version 3 of the GNU Affero General Public License. 64 | 65 | "Copyright" also means copyright-like laws that apply to other kinds of 66 | works, such as semiconductor masks. 67 | 68 | "The Program" refers to any copyrightable work licensed under this 69 | License. Each licensee is addressed as "you". "Licensees" and 70 | "recipients" may be individuals or organizations. 71 | 72 | To "modify" a work means to copy from or adapt all or part of the work 73 | in a fashion requiring copyright permission, other than the making of an 74 | exact copy. The resulting work is called a "modified version" of the 75 | earlier work or a work "based on" the earlier work. 76 | 77 | A "covered work" means either the unmodified Program or a work based 78 | on the Program. 79 | 80 | To "propagate" a work means to do anything with it that, without 81 | permission, would make you directly or secondarily liable for 82 | infringement under applicable copyright law, except executing it on a 83 | computer or modifying a private copy. Propagation includes copying, 84 | distribution (with or without modification), making available to the 85 | public, and in some countries other activities as well. 86 | 87 | To "convey" a work means any kind of propagation that enables other 88 | parties to make or receive copies. Mere interaction with a user through 89 | a computer network, with no transfer of a copy, is not conveying. 90 | 91 | An interactive user interface displays "Appropriate Legal Notices" 92 | to the extent that it includes a convenient and prominently visible 93 | feature that (1) displays an appropriate copyright notice, and (2) 94 | tells the user that there is no warranty for the work (except to the 95 | extent that warranties are provided), that licensees may convey the 96 | work under this License, and how to view a copy of this License. If 97 | the interface presents a list of user commands or options, such as a 98 | menu, a prominent item in the list meets this criterion. 99 | 100 | 1. Source Code. 101 | 102 | The "source code" for a work means the preferred form of the work 103 | for making modifications to it. "Object code" means any non-source 104 | form of a work. 105 | 106 | A "Standard Interface" means an interface that either is an official 107 | standard defined by a recognized standards body, or, in the case of 108 | interfaces specified for a particular programming language, one that 109 | is widely used among developers working in that language. 110 | 111 | The "System Libraries" of an executable work include anything, other 112 | than the work as a whole, that (a) is included in the normal form of 113 | packaging a Major Component, but which is not part of that Major 114 | Component, and (b) serves only to enable use of the work with that 115 | Major Component, or to implement a Standard Interface for which an 116 | implementation is available to the public in source code form. A 117 | "Major Component", in this context, means a major essential component 118 | (kernel, window system, and so on) of the specific operating system 119 | (if any) on which the executable work runs, or a compiler used to 120 | produce the work, or an object code interpreter used to run it. 121 | 122 | The "Corresponding Source" for a work in object code form means all 123 | the source code needed to generate, install, and (for an executable 124 | work) run the object code and to modify the work, including scripts to 125 | control those activities. However, it does not include the work's 126 | System Libraries, or general-purpose tools or generally available free 127 | programs which are used unmodified in performing those activities but 128 | which are not part of the work. For example, Corresponding Source 129 | includes interface definition files associated with source files for 130 | the work, and the source code for shared libraries and dynamically 131 | linked subprograms that the work is specifically designed to require, 132 | such as by intimate data communication or control flow between those 133 | subprograms and other parts of the work. 134 | 135 | The Corresponding Source need not include anything that users 136 | can regenerate automatically from other parts of the Corresponding 137 | Source. 138 | 139 | The Corresponding Source for a work in source code form is that 140 | same work. 141 | 142 | 2. Basic Permissions. 143 | 144 | All rights granted under this License are granted for the term of 145 | copyright on the Program, and are irrevocable provided the stated 146 | conditions are met. This License explicitly affirms your unlimited 147 | permission to run the unmodified Program. The output from running a 148 | covered work is covered by this License only if the output, given its 149 | content, constitutes a covered work. This License acknowledges your 150 | rights of fair use or other equivalent, as provided by copyright law. 151 | 152 | You may make, run and propagate covered works that you do not 153 | convey, without conditions so long as your license otherwise remains 154 | in force. You may convey covered works to others for the sole purpose 155 | of having them make modifications exclusively for you, or provide you 156 | with facilities for running those works, provided that you comply with 157 | the terms of this License in conveying all material for which you do 158 | not control copyright. Those thus making or running the covered works 159 | for you must do so exclusively on your behalf, under your direction 160 | and control, on terms that prohibit them from making any copies of 161 | your copyrighted material outside their relationship with you. 162 | 163 | Conveying under any other circumstances is permitted solely under 164 | the conditions stated below. Sublicensing is not allowed; section 10 165 | makes it unnecessary. 166 | 167 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 168 | 169 | No covered work shall be deemed part of an effective technological 170 | measure under any applicable law fulfilling obligations under article 171 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or 172 | similar laws prohibiting or restricting circumvention of such 173 | measures. 174 | 175 | When you convey a covered work, you waive any legal power to forbid 176 | circumvention of technological measures to the extent such circumvention 177 | is effected by exercising rights under this License with respect to 178 | the covered work, and you disclaim any intention to limit operation or 179 | modification of the work as a means of enforcing, against the work's 180 | users, your or third parties' legal rights to forbid circumvention of 181 | technological measures. 182 | 183 | 4. Conveying Verbatim Copies. 184 | 185 | You may convey verbatim copies of the Program's source code as you 186 | receive it, in any medium, provided that you conspicuously and 187 | appropriately publish on each copy an appropriate copyright notice; 188 | keep intact all notices stating that this License and any 189 | non-permissive terms added in accord with section 7 apply to the code; 190 | keep intact all notices of the absence of any warranty; and give all 191 | recipients a copy of this License along with the Program. 192 | 193 | You may charge any price or no price for each copy that you convey, 194 | and you may offer support or warranty protection for a fee. 195 | 196 | 5. Conveying Modified Source Versions. 197 | 198 | You may convey a work based on the Program, or the modifications to 199 | produce it from the Program, in the form of source code under the 200 | terms of section 4, provided that you also meet all of these conditions: 201 | 202 | a) The work must carry prominent notices stating that you modified 203 | it, and giving a relevant date. 204 | 205 | b) The work must carry prominent notices stating that it is 206 | released under this License and any conditions added under section 207 | 7. This requirement modifies the requirement in section 4 to 208 | "keep intact all notices". 209 | 210 | c) You must license the entire work, as a whole, under this 211 | License to anyone who comes into possession of a copy. This 212 | License will therefore apply, along with any applicable section 7 213 | additional terms, to the whole of the work, and all its parts, 214 | regardless of how they are packaged. This License gives no 215 | permission to license the work in any other way, but it does not 216 | invalidate such permission if you have separately received it. 217 | 218 | d) If the work has interactive user interfaces, each must display 219 | Appropriate Legal Notices; however, if the Program has interactive 220 | interfaces that do not display Appropriate Legal Notices, your 221 | work need not make them do so. 222 | 223 | A compilation of a covered work with other separate and independent 224 | works, which are not by their nature extensions of the covered work, 225 | and which are not combined with it such as to form a larger program, 226 | in or on a volume of a storage or distribution medium, is called an 227 | "aggregate" if the compilation and its resulting copyright are not 228 | used to limit the access or legal rights of the compilation's users 229 | beyond what the individual works permit. Inclusion of a covered work 230 | in an aggregate does not cause this License to apply to the other 231 | parts of the aggregate. 232 | 233 | 6. Conveying Non-Source Forms. 234 | 235 | You may convey a covered work in object code form under the terms 236 | of sections 4 and 5, provided that you also convey the 237 | machine-readable Corresponding Source under the terms of this License, 238 | in one of these ways: 239 | 240 | a) Convey the object code in, or embodied in, a physical product 241 | (including a physical distribution medium), accompanied by the 242 | Corresponding Source fixed on a durable physical medium 243 | customarily used for software interchange. 244 | 245 | b) Convey the object code in, or embodied in, a physical product 246 | (including a physical distribution medium), accompanied by a 247 | written offer, valid for at least three years and valid for as 248 | long as you offer spare parts or customer support for that product 249 | model, to give anyone who possesses the object code either (1) a 250 | copy of the Corresponding Source for all the software in the 251 | product that is covered by this License, on a durable physical 252 | medium customarily used for software interchange, for a price no 253 | more than your reasonable cost of physically performing this 254 | conveying of source, or (2) access to copy the 255 | Corresponding Source from a network server at no charge. 256 | 257 | c) Convey individual copies of the object code with a copy of the 258 | written offer to provide the Corresponding Source. This 259 | alternative is allowed only occasionally and noncommercially, and 260 | only if you received the object code with such an offer, in accord 261 | with subsection 6b. 262 | 263 | d) Convey the object code by offering access from a designated 264 | place (gratis or for a charge), and offer equivalent access to the 265 | Corresponding Source in the same way through the same place at no 266 | further charge. You need not require recipients to copy the 267 | Corresponding Source along with the object code. If the place to 268 | copy the object code is a network server, the Corresponding Source 269 | may be on a different server (operated by you or a third party) 270 | that supports equivalent copying facilities, provided you maintain 271 | clear directions next to the object code saying where to find the 272 | Corresponding Source. Regardless of what server hosts the 273 | Corresponding Source, you remain obligated to ensure that it is 274 | available for as long as needed to satisfy these requirements. 275 | 276 | e) Convey the object code using peer-to-peer transmission, provided 277 | you inform other peers where the object code and Corresponding 278 | Source of the work are being offered to the general public at no 279 | charge under subsection 6d. 280 | 281 | A separable portion of the object code, whose source code is excluded 282 | from the Corresponding Source as a System Library, need not be 283 | included in conveying the object code work. 284 | 285 | A "User Product" is either (1) a "consumer product", which means any 286 | tangible personal property which is normally used for personal, family, 287 | or household purposes, or (2) anything designed or sold for incorporation 288 | into a dwelling. In determining whether a product is a consumer product, 289 | doubtful cases shall be resolved in favor of coverage. For a particular 290 | product received by a particular user, "normally used" refers to a 291 | typical or common use of that class of product, regardless of the status 292 | of the particular user or of the way in which the particular user 293 | actually uses, or expects or is expected to use, the product. A product 294 | is a consumer product regardless of whether the product has substantial 295 | commercial, industrial or non-consumer uses, unless such uses represent 296 | the only significant mode of use of the product. 297 | 298 | "Installation Information" for a User Product means any methods, 299 | procedures, authorization keys, or other information required to install 300 | and execute modified versions of a covered work in that User Product from 301 | a modified version of its Corresponding Source. The information must 302 | suffice to ensure that the continued functioning of the modified object 303 | code is in no case prevented or interfered with solely because 304 | modification has been made. 305 | 306 | If you convey an object code work under this section in, or with, or 307 | specifically for use in, a User Product, and the conveying occurs as 308 | part of a transaction in which the right of possession and use of the 309 | User Product is transferred to the recipient in perpetuity or for a 310 | fixed term (regardless of how the transaction is characterized), the 311 | Corresponding Source conveyed under this section must be accompanied 312 | by the Installation Information. But this requirement does not apply 313 | if neither you nor any third party retains the ability to install 314 | modified object code on the User Product (for example, the work has 315 | been installed in ROM). 316 | 317 | The requirement to provide Installation Information does not include a 318 | requirement to continue to provide support service, warranty, or updates 319 | for a work that has been modified or installed by the recipient, or for 320 | the User Product in which it has been modified or installed. Access to a 321 | network may be denied when the modification itself materially and 322 | adversely affects the operation of the network or violates the rules and 323 | protocols for communication across the network. 324 | 325 | Corresponding Source conveyed, and Installation Information provided, 326 | in accord with this section must be in a format that is publicly 327 | documented (and with an implementation available to the public in 328 | source code form), and must require no special password or key for 329 | unpacking, reading or copying. 330 | 331 | 7. Additional Terms. 332 | 333 | "Additional permissions" are terms that supplement the terms of this 334 | License by making exceptions from one or more of its conditions. 335 | Additional permissions that are applicable to the entire Program shall 336 | be treated as though they were included in this License, to the extent 337 | that they are valid under applicable law. If additional permissions 338 | apply only to part of the Program, that part may be used separately 339 | under those permissions, but the entire Program remains governed by 340 | this License without regard to the additional permissions. 341 | 342 | When you convey a copy of a covered work, you may at your option 343 | remove any additional permissions from that copy, or from any part of 344 | it. (Additional permissions may be written to require their own 345 | removal in certain cases when you modify the work.) You may place 346 | additional permissions on material, added by you to a covered work, 347 | for which you have or can give appropriate copyright permission. 348 | 349 | Notwithstanding any other provision of this License, for material you 350 | add to a covered work, you may (if authorized by the copyright holders of 351 | that material) supplement the terms of this License with terms: 352 | 353 | a) Disclaiming warranty or limiting liability differently from the 354 | terms of sections 15 and 16 of this License; or 355 | 356 | b) Requiring preservation of specified reasonable legal notices or 357 | author attributions in that material or in the Appropriate Legal 358 | Notices displayed by works containing it; or 359 | 360 | c) Prohibiting misrepresentation of the origin of that material, or 361 | requiring that modified versions of such material be marked in 362 | reasonable ways as different from the original version; or 363 | 364 | d) Limiting the use for publicity purposes of names of licensors or 365 | authors of the material; or 366 | 367 | e) Declining to grant rights under trademark law for use of some 368 | trade names, trademarks, or service marks; or 369 | 370 | f) Requiring indemnification of licensors and authors of that 371 | material by anyone who conveys the material (or modified versions of 372 | it) with contractual assumptions of liability to the recipient, for 373 | any liability that these contractual assumptions directly impose on 374 | those licensors and authors. 375 | 376 | All other non-permissive additional terms are considered "further 377 | restrictions" within the meaning of section 10. If the Program as you 378 | received it, or any part of it, contains a notice stating that it is 379 | governed by this License along with a term that is a further 380 | restriction, you may remove that term. If a license document contains 381 | a further restriction but permits relicensing or conveying under this 382 | License, you may add to a covered work material governed by the terms 383 | of that license document, provided that the further restriction does 384 | not survive such relicensing or conveying. 385 | 386 | If you add terms to a covered work in accord with this section, you 387 | must place, in the relevant source files, a statement of the 388 | additional terms that apply to those files, or a notice indicating 389 | where to find the applicable terms. 390 | 391 | Additional terms, permissive or non-permissive, may be stated in the 392 | form of a separately written license, or stated as exceptions; 393 | the above requirements apply either way. 394 | 395 | 8. Termination. 396 | 397 | You may not propagate or modify a covered work except as expressly 398 | provided under this License. Any attempt otherwise to propagate or 399 | modify it is void, and will automatically terminate your rights under 400 | this License (including any patent licenses granted under the third 401 | paragraph of section 11). 402 | 403 | However, if you cease all violation of this License, then your 404 | license from a particular copyright holder is reinstated (a) 405 | provisionally, unless and until the copyright holder explicitly and 406 | finally terminates your license, and (b) permanently, if the copyright 407 | holder fails to notify you of the violation by some reasonable means 408 | prior to 60 days after the cessation. 409 | 410 | Moreover, your license from a particular copyright holder is 411 | reinstated permanently if the copyright holder notifies you of the 412 | violation by some reasonable means, this is the first time you have 413 | received notice of violation of this License (for any work) from that 414 | copyright holder, and you cure the violation prior to 30 days after 415 | your receipt of the notice. 416 | 417 | Termination of your rights under this section does not terminate the 418 | licenses of parties who have received copies or rights from you under 419 | this License. If your rights have been terminated and not permanently 420 | reinstated, you do not qualify to receive new licenses for the same 421 | material under section 10. 422 | 423 | 9. Acceptance Not Required for Having Copies. 424 | 425 | You are not required to accept this License in order to receive or 426 | run a copy of the Program. Ancillary propagation of a covered work 427 | occurring solely as a consequence of using peer-to-peer transmission 428 | to receive a copy likewise does not require acceptance. However, 429 | nothing other than this License grants you permission to propagate or 430 | modify any covered work. These actions infringe copyright if you do 431 | not accept this License. Therefore, by modifying or propagating a 432 | covered work, you indicate your acceptance of this License to do so. 433 | 434 | 10. Automatic Licensing of Downstream Recipients. 435 | 436 | Each time you convey a covered work, the recipient automatically 437 | receives a license from the original licensors, to run, modify and 438 | propagate that work, subject to this License. You are not responsible 439 | for enforcing compliance by third parties with this License. 440 | 441 | An "entity transaction" is a transaction transferring control of an 442 | organization, or substantially all assets of one, or subdividing an 443 | organization, or merging organizations. If propagation of a covered 444 | work results from an entity transaction, each party to that 445 | transaction who receives a copy of the work also receives whatever 446 | licenses to the work the party's predecessor in interest had or could 447 | give under the previous paragraph, plus a right to possession of the 448 | Corresponding Source of the work from the predecessor in interest, if 449 | the predecessor has it or can get it with reasonable efforts. 450 | 451 | You may not impose any further restrictions on the exercise of the 452 | rights granted or affirmed under this License. For example, you may 453 | not impose a license fee, royalty, or other charge for exercise of 454 | rights granted under this License, and you may not initiate litigation 455 | (including a cross-claim or counterclaim in a lawsuit) alleging that 456 | any patent claim is infringed by making, using, selling, offering for 457 | sale, or importing the Program or any portion of it. 458 | 459 | 11. Patents. 460 | 461 | A "contributor" is a copyright holder who authorizes use under this 462 | License of the Program or a work on which the Program is based. The 463 | work thus licensed is called the contributor's "contributor version". 464 | 465 | A contributor's "essential patent claims" are all patent claims 466 | owned or controlled by the contributor, whether already acquired or 467 | hereafter acquired, that would be infringed by some manner, permitted 468 | by this License, of making, using, or selling its contributor version, 469 | but do not include claims that would be infringed only as a 470 | consequence of further modification of the contributor version. For 471 | purposes of this definition, "control" includes the right to grant 472 | patent sublicenses in a manner consistent with the requirements of 473 | this License. 474 | 475 | Each contributor grants you a non-exclusive, worldwide, royalty-free 476 | patent license under the contributor's essential patent claims, to 477 | make, use, sell, offer for sale, import and otherwise run, modify and 478 | propagate the contents of its contributor version. 479 | 480 | In the following three paragraphs, a "patent license" is any express 481 | agreement or commitment, however denominated, not to enforce a patent 482 | (such as an express permission to practice a patent or covenant not to 483 | sue for patent infringement). To "grant" such a patent license to a 484 | party means to make such an agreement or commitment not to enforce a 485 | patent against the party. 486 | 487 | If you convey a covered work, knowingly relying on a patent license, 488 | and the Corresponding Source of the work is not available for anyone 489 | to copy, free of charge and under the terms of this License, through a 490 | publicly available network server or other readily accessible means, 491 | then you must either (1) cause the Corresponding Source to be so 492 | available, or (2) arrange to deprive yourself of the benefit of the 493 | patent license for this particular work, or (3) arrange, in a manner 494 | consistent with the requirements of this License, to extend the patent 495 | license to downstream recipients. "Knowingly relying" means you have 496 | actual knowledge that, but for the patent license, your conveying the 497 | covered work in a country, or your recipient's use of the covered work 498 | in a country, would infringe one or more identifiable patents in that 499 | country that you have reason to believe are valid. 500 | 501 | If, pursuant to or in connection with a single transaction or 502 | arrangement, you convey, or propagate by procuring conveyance of, a 503 | covered work, and grant a patent license to some of the parties 504 | receiving the covered work authorizing them to use, propagate, modify 505 | or convey a specific copy of the covered work, then the patent license 506 | you grant is automatically extended to all recipients of the covered 507 | work and works based on it. 508 | 509 | A patent license is "discriminatory" if it does not include within 510 | the scope of its coverage, prohibits the exercise of, or is 511 | conditioned on the non-exercise of one or more of the rights that are 512 | specifically granted under this License. You may not convey a covered 513 | work if you are a party to an arrangement with a third party that is 514 | in the business of distributing software, under which you make payment 515 | to the third party based on the extent of your activity of conveying 516 | the work, and under which the third party grants, to any of the 517 | parties who would receive the covered work from you, a discriminatory 518 | patent license (a) in connection with copies of the covered work 519 | conveyed by you (or copies made from those copies), or (b) primarily 520 | for and in connection with specific products or compilations that 521 | contain the covered work, unless you entered into that arrangement, 522 | or that patent license was granted, prior to 28 March 2007. 523 | 524 | Nothing in this License shall be construed as excluding or limiting 525 | any implied license or other defenses to infringement that may 526 | otherwise be available to you under applicable patent law. 527 | 528 | 12. No Surrender of Others' Freedom. 529 | 530 | If conditions are imposed on you (whether by court order, agreement or 531 | otherwise) that contradict the conditions of this License, they do not 532 | excuse you from the conditions of this License. If you cannot convey a 533 | covered work so as to satisfy simultaneously your obligations under this 534 | License and any other pertinent obligations, then as a consequence you may 535 | not convey it at all. For example, if you agree to terms that obligate you 536 | to collect a royalty for further conveying from those to whom you convey 537 | the Program, the only way you could satisfy both those terms and this 538 | License would be to refrain entirely from conveying the Program. 539 | 540 | 13. Remote Network Interaction; Use with the GNU General Public License. 541 | 542 | Notwithstanding any other provision of this License, if you modify the 543 | Program, your modified version must prominently offer all users 544 | interacting with it remotely through a computer network (if your version 545 | supports such interaction) an opportunity to receive the Corresponding 546 | Source of your version by providing access to the Corresponding Source 547 | from a network server at no charge, through some standard or customary 548 | means of facilitating copying of software. This Corresponding Source 549 | shall include the Corresponding Source for any work covered by version 3 550 | of the GNU General Public License that is incorporated pursuant to the 551 | following paragraph. 552 | 553 | Notwithstanding any other provision of this License, you have 554 | permission to link or combine any covered work with a work licensed 555 | under version 3 of the GNU General Public License into a single 556 | combined work, and to convey the resulting work. The terms of this 557 | License will continue to apply to the part which is the covered work, 558 | but the work with which it is combined will remain governed by version 559 | 3 of the GNU General Public License. 560 | 561 | 14. Revised Versions of this License. 562 | 563 | The Free Software Foundation may publish revised and/or new versions of 564 | the GNU Affero General Public License from time to time. Such new versions 565 | will be similar in spirit to the present version, but may differ in detail to 566 | address new problems or concerns. 567 | 568 | Each version is given a distinguishing version number. If the 569 | Program specifies that a certain numbered version of the GNU Affero General 570 | Public License "or any later version" applies to it, you have the 571 | option of following the terms and conditions either of that numbered 572 | version or of any later version published by the Free Software 573 | Foundation. If the Program does not specify a version number of the 574 | GNU Affero General Public License, you may choose any version ever published 575 | by the Free Software Foundation. 576 | 577 | If the Program specifies that a proxy can decide which future 578 | versions of the GNU Affero General Public License can be used, that proxy's 579 | public statement of acceptance of a version permanently authorizes you 580 | to choose that version for the Program. 581 | 582 | Later license versions may give you additional or different 583 | permissions. However, no additional obligations are imposed on any 584 | author or copyright holder as a result of your choosing to follow a 585 | later version. 586 | 587 | 15. Disclaimer of Warranty. 588 | 589 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 590 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT 591 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY 592 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, 593 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 594 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM 595 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF 596 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 597 | 598 | 16. Limitation of Liability. 599 | 600 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING 601 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS 602 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY 603 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE 604 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF 605 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 606 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), 607 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF 608 | SUCH DAMAGES. 609 | 610 | 17. Interpretation of Sections 15 and 16. 611 | 612 | If the disclaimer of warranty and limitation of liability provided 613 | above cannot be given local legal effect according to their terms, 614 | reviewing courts shall apply local law that most closely approximates 615 | an absolute waiver of all civil liability in connection with the 616 | Program, unless a warranty or assumption of liability accompanies a 617 | copy of the Program in return for a fee. 618 | 619 | END OF TERMS AND CONDITIONS 620 | 621 | How to Apply These Terms to Your New Programs 622 | 623 | If you develop a new program, and you want it to be of the greatest 624 | possible use to the public, the best way to achieve this is to make it 625 | free software which everyone can redistribute and change under these terms. 626 | 627 | To do so, attach the following notices to the program. It is safest 628 | to attach them to the start of each source file to most effectively 629 | state the exclusion of warranty; and each file should have at least 630 | the "copyright" line and a pointer to where the full notice is found. 631 | 632 | 633 | Copyright (C) 634 | 635 | This program is free software: you can redistribute it and/or modify 636 | it under the terms of the GNU Affero General Public License as published 637 | by the Free Software Foundation, either version 3 of the License, or 638 | (at your option) any later version. 639 | 640 | This program is distributed in the hope that it will be useful, 641 | but WITHOUT ANY WARRANTY; without even the implied warranty of 642 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 643 | GNU Affero General Public License for more details. 644 | 645 | You should have received a copy of the GNU Affero General Public License 646 | along with this program. If not, see . 647 | 648 | Also add information on how to contact you by electronic and paper mail. 649 | 650 | If your software can interact with users remotely through a computer 651 | network, you should also make sure that it provides a way for users to 652 | get its source. For example, if your program is a web application, its 653 | interface could display a "Source" link that leads users to an archive 654 | of the code. There are many ways you could offer source, and different 655 | solutions will be better for different programs; see section 13 for the 656 | specific requirements. 657 | 658 | You should also get your employer (if you work as a programmer) or school, 659 | if any, to sign a "copyright disclaimer" for the program, if necessary. 660 | For more information on this, and how to apply and follow the GNU AGPL, see 661 | . 662 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Audiocraft Infinity WebUI 2 | 3 | Adds generation of songs with a length of over 30 seconds. 4 | 5 | Adds the ability to continue songs. 6 | 7 | Adds a seed option. 8 | 9 | Adds ability to load locally downloaded models. 10 | 11 | ### Adds training (Thanks to chavinlo's repo https://github.com/chavinlo/musicgen_trainer) 12 | 13 | Adds MacOS support. 14 | 15 | Adds queue (on the main-queue branch: https://github.com/1aienthusiast/audiocraft-infinity-webui/tree/main-queue) 16 | 17 | Batching (**run webuibatch.py instead of webui.py**) 18 | 19 | Disables (hopefully) the gradio analytics. 20 | 21 | ## Note! Project is currently not actively maintained but accepts PRs. 22 | 23 | ## Installation 24 | Python 3.9 is recommended. 25 | 26 | 1. Clone the repo: 27 | `git clone https://github.com/1aienthusiast/audiocraft-infinity-webui.git` 28 | 2. Install pytorch: 29 | `pip install 'torch>=2.0'` 30 | 3. Install the requirements: 31 | `pip install -r requirements.txt` 32 | 4. Clone my fork of the Meta audiocraft repo and chavinlo's MusicGen trainer inside the `repositories` folder: 33 | ``` 34 | cd repositories 35 | git clone https://github.com/1aienthusiast/audiocraft 36 | git clone https://github.com/chavinlo/musicgen_trainer 37 | cd .. 38 | ``` 39 | ## Note! 40 | If you already cloned the Meta audiocraft repo you have to remove it then clone the provided fork for the seed option to work. 41 | ``` 42 | cd repositories 43 | rm -rf audiocraft/ 44 | git clone https://github.com/1aienthusiast/audiocraft 45 | git clone https://github.com/chavinlo/musicgen_trainer 46 | cd .. 47 | ``` 48 | 49 | ## Usage 50 | ```python webui.py``` 51 | ```python webuibatch.py``` - with batching support 52 | 53 | ## Updating 54 | Run `git pull` inside the root folder to update the webui, and the same command inside `repositories/audiocraft` to update audiocraft. 55 | 56 | ## Models 57 | 58 | Meta provides 4 pre-trained models. The pre trained models are: 59 | - `small`: 300M model, text to music only - [🤗 Hub](https://huggingface.co/facebook/musicgen-small) 60 | - `medium`: 1.5B model, text to music only - [🤗 Hub](https://huggingface.co/facebook/musicgen-medium) 61 | - `melody`: 1.5B model, text to music and text+melody to music - [🤗 Hub](https://huggingface.co/facebook/musicgen-melody) 62 | - `large`: 3.3B model, text to music only - [🤗 Hub](https://huggingface.co/facebook/musicgen-large) 63 | 64 | **Needs a GPU!** 65 | 66 | I recommend 12GB of VRAM for the large model. 67 | 68 | ## Training 69 | 70 | ### Dataset Creation 71 | 72 | Create a folder, in it, place your audio and caption files. **They must be WAV and TXT format respectively.** 73 | 74 | ![](https://i.imgur.com/AlDlqBI.png) 75 | 76 | **Place the folder in `training/datasets/`.** 77 | 78 | ### Important: Split your audios in 35 second chunks. Only the first 30 seconds will be processed. Audio cannot be less than 30 seconds. 79 | 80 | In this example, segment_000.txt contains the caption "jazz music, jobim" for wav file segment_000.wav 81 | 82 | ### Options 83 | 84 | - `dataset_path` - path to your dataset with WAV and TXT pairs. 85 | - `model_id` - MusicGen model to use. Can be `small`/`medium`/`large`. Default: `small` - model it will be finetuned on 86 | - `lr`: Float, learning rate. Default: `0.0001`/`1e-4` 87 | - `epochs`: Integer, epoch count. Default: `5` 88 | - `use_wandb`: Integer, `1` to enable wandb, `0` to disable it. Default: `0` = Disabled 89 | - `save_step`: Integer, amount of steps to save a checkpoint. Default: None 90 | 91 | ### Models 92 | 93 | Once training finishes, the model (and checkpoints) will be available under the `models/` directory. 94 | 95 | ![](https://i.imgur.com/Mu19EPb.png) 96 | 97 | ### Loading the finetuned models 98 | Model gets saved to models/ as `lm_final.pt` 99 | 100 | 1) Place it in models/DIRECTORY_NAME/ 101 | 2) In the Inference tab choose `custom` as the model and enter DIRECTORY_NAME into the input field. 102 | 3) In the Inference tab choose the model it was finetuned on 103 | 104 | ## Colab 105 | 106 | For google colab you need to use the `--share` flag. 107 | 108 | ## License 109 | * The code in this repository is released under the AGPLv3 license as found in the [LICENSE file](LICENSE). 110 | -------------------------------------------------------------------------------- /models/place-models-here.txt: -------------------------------------------------------------------------------- 1 | Place models here, if they are the original ones from meta they must have same names as on huggingface. 2 | -------------------------------------------------------------------------------- /modules/shared.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | 3 | parser = argparse.ArgumentParser(formatter_class=lambda prog: argparse.HelpFormatter(prog, max_help_position=54)) 4 | 5 | parser.add_argument('--listen', action='store_true', help='Make the web UI reachable from your local network.') 6 | parser.add_argument('--listen-host', type=str, help='The hostname that the server will use.') 7 | parser.add_argument('--listen-port', type=int, help='The listening port that the server will use.') 8 | parser.add_argument('--share', action='store_true', help='Create a public URL. This is useful for running the web UI on Google Colab or similar.') 9 | parser.add_argument('--auto-launch', action='store_true', default=False, help='Open the web UI in the default browser upon launch.') 10 | 11 | args = parser.parse_args() 12 | -------------------------------------------------------------------------------- /modules/ui.py: -------------------------------------------------------------------------------- 1 | import requests 2 | import os 3 | os.environ['GRADIO_ANALYTICS_ENABLED'] = 'False' 4 | def my_get(url, **kwargs): 5 | kwargs.setdefault('allow_redirects', True) 6 | return requests.api.request('get', 'http://127.0.0.1/', **kwargs) 7 | original_get = requests.get 8 | requests.get = my_get 9 | import gradio as gr 10 | requests.get = original_get 11 | refresh_symbol = '\U0001f504' 12 | 13 | class ToolButton(gr.Button, gr.components.FormComponent): 14 | """Small button with single emoji as text, fits inside gradio forms""" 15 | 16 | def __init__(self, **kwargs): 17 | super().__init__(**kwargs) 18 | 19 | def get_block_name(self): 20 | return "button" 21 | 22 | def create_refresh_button(refresh_component, refresh_method, refreshed_args, elem_class): 23 | def refresh(): 24 | refresh_method() 25 | args = refreshed_args() if callable(refreshed_args) else refreshed_args 26 | 27 | for k, v in args.items(): 28 | setattr(refresh_component, k, v) 29 | 30 | return gr.update(**(args or {})) 31 | 32 | refresh_button = ToolButton(value=refresh_symbol, elem_classes=elem_class) 33 | refresh_button.click( 34 | fn=refresh, 35 | inputs=[], 36 | outputs=[refresh_component] 37 | ) 38 | return refresh_button 39 | -------------------------------------------------------------------------------- /repositories/clone-audiocraft-here.txt: -------------------------------------------------------------------------------- 1 | 2 | -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | # please make sure you have already a pytorch install that is cuda enabled! 2 | av 3 | einops 4 | flashy>=0.0.1 5 | hydra-core>=1.1 6 | hydra_colorlog 7 | julius 8 | num2words 9 | numpy 10 | sentencepiece 11 | spacy==3.5.2 12 | torch>=2.0.0 13 | torchaudio>=2.0.0 14 | huggingface_hub 15 | tqdm 16 | temp 17 | transformers 18 | xformers 19 | demucs 20 | librosa 21 | requests 22 | gradio==3.34.0 23 | contextlib2 24 | wave 25 | wandb 26 | -------------------------------------------------------------------------------- /results/results-get-saved-here.txt: -------------------------------------------------------------------------------- 1 | Results will be saved here 2 | -------------------------------------------------------------------------------- /training/datasets/place-datasets-here.md: -------------------------------------------------------------------------------- 1 | ### Dataset Creation 2 | 3 | Create a folder, in it, place your audio and caption files. **They must be WAV and TXT format respectively.** 4 | 5 | ![](https://i.imgur.com/AlDlqBI.png) 6 | -------------------------------------------------------------------------------- /webui.py: -------------------------------------------------------------------------------- 1 | import contextlib 2 | from datetime import datetime 3 | import os 4 | import random 5 | import sys 6 | import typing 7 | import wave 8 | import glob 9 | from pathlib import Path 10 | from tempfile import NamedTemporaryFile 11 | 12 | import numpy as np 13 | import requests 14 | import torch 15 | import torchaudio 16 | from os.path import dirname, abspath 17 | from modules import shared, ui 18 | cuda = True 19 | #check for mps 20 | if torch.backends.mps.is_available(): 21 | mps_device = torch.device("mps") 22 | x = torch.ones(1, device=mps_device) 23 | cuda = False 24 | 25 | sys.path.insert(0, str(Path("repositories/audiocraft"))) 26 | sys.path.insert(0, str(Path("repositories/musicgen_trainer"))) 27 | from train import train 28 | from audiocraft.data.audio import audio_write 29 | from audiocraft.data.audio_utils import convert_audio 30 | from audiocraft.models import MusicGen 31 | 32 | os.environ['GRADIO_ANALYTICS_ENABLED'] = 'False' 33 | def my_get(url, **kwargs): 34 | kwargs.setdefault('allow_redirects', True) 35 | return requests.api.request('get', 'http://127.0.0.1/', **kwargs) 36 | original_get = requests.get 37 | requests.get = my_get 38 | import gradio as gr 39 | requests.get = original_get 40 | 41 | 42 | MODEL = None 43 | current_directory = dirname(abspath(__file__)) 44 | 45 | def load_model(version, DIRECTORY_NAME, FINETUNED_ON): 46 | if version != "custom": 47 | print("Loading model", version) 48 | path=current_directory+"/models/" + version + "/" 49 | if os.path.exists(path): 50 | model = MusicGen.get_pretrained(directory=path,name=version) 51 | 52 | else: model = MusicGen.get_pretrained(name=version) 53 | else: 54 | finetuned_dir =current_directory + "/models/" + DIRECTORY_NAME + "/" + "lm_final.pt" 55 | model= MusicGen.get_pretrained(name=FINETUNED_ON) 56 | model.lm.load_state_dict(torch.load(finetuned_dir)) 57 | model.name="custom" 58 | return model 59 | 60 | 61 | def set_seed(seed: int = 0): 62 | torch.backends.cudnn.deterministic = True 63 | torch.backends.cudnn.benchmark = False 64 | if seed <= 0: 65 | seed = np.random.default_rng().integers(1, 2**32 - 1) 66 | seed = np.uint32(seed).item() 67 | assert 0 < seed < 2**32 68 | original_seed = seed 69 | np.random.seed(seed) 70 | random.seed(seed) 71 | torch.manual_seed(seed) 72 | if cuda: 73 | torch.cuda.manual_seed_all(seed) 74 | os.environ["PYTHONHASHSEED"] = str(seed) 75 | return original_seed 76 | 77 | 78 | def generate_cmelody(descriptions: typing.List[str], melody_wavs: typing.Union[torch.Tensor, typing.List[typing.Optional[torch.Tensor]]], 79 | msr: int, prompt: torch.Tensor, psr: int, MODEL, progress: bool = False) -> torch.Tensor: 80 | if isinstance(melody_wavs, torch.Tensor): 81 | if melody_wavs.dim() == 2: 82 | melody_wavs = melody_wavs[None] 83 | if melody_wavs.dim() != 3: 84 | raise ValueError("melody_wavs should have a shape [B, C, T].") 85 | melody_wavs = list(melody_wavs) 86 | else: 87 | for melody in melody_wavs: 88 | if melody is not None: 89 | assert melody.dim() == 2, "one melody in the list has the wrong number of dims." 90 | 91 | melody_wavs = [ 92 | convert_audio(wav, msr, MODEL.sample_rate, MODEL.audio_channels) 93 | if wav is not None else None 94 | for wav in melody_wavs] 95 | 96 | if prompt.dim() == 2: 97 | prompt = prompt[None] 98 | if prompt.dim() != 3: 99 | raise ValueError("prompt should have 3 dimensions: [B, C, T] (C = 1).") 100 | prompt = convert_audio(prompt, psr, MODEL.sample_rate, MODEL.audio_channels) 101 | if descriptions is None: 102 | descriptions = [None] * len(prompt) 103 | attributes, prompt_tokens = MusicGen._prepare_tokens_and_attributes(MODEL, descriptions=descriptions, prompt=prompt, melody_wavs=melody_wavs) 104 | assert prompt_tokens is not None 105 | return MusicGen._generate_tokens(MODEL, attributes, prompt_tokens, progress) 106 | 107 | 108 | def initial_generate(melody_boolean, MODEL, text, melody, msr, continue_file, duration, cf_cutoff, sc_text): 109 | wav = None 110 | if continue_file: 111 | data_waveform, cfsr = (torchaudio.load(continue_file)) 112 | if cuda: 113 | wav = data_waveform.cuda() 114 | else: 115 | wav = data_waveform.mps_device() 116 | cf_len = 0 117 | with contextlib.closing(wave.open(continue_file, 'r')) as f: 118 | frames = f.getnframes() 119 | rate = f.getframerate() 120 | cf_len = frames / float(rate) 121 | if wav.dim() == 2: 122 | wav = wav[None] 123 | wav = wav[:, :, int(-cfsr * min(29, cf_len, duration - 1, cf_cutoff)):] 124 | new_chunk = None 125 | if not melody_boolean: 126 | if not sc_text: 127 | new_chunk = MODEL.generate_continuation(wav, prompt_sample_rate=cfsr, progress=False) 128 | else: 129 | new_chunk = MODEL.generate_continuation(wav, descriptions=[text], prompt_sample_rate=cfsr, progress=False) 130 | wav = new_chunk 131 | else: 132 | new_chunk = generate_cmelody([text], melody, msr, wav, cfsr, MODEL, progress=False) 133 | wav = new_chunk 134 | else: 135 | if melody_boolean: 136 | wav = MODEL.generate_with_chroma( 137 | descriptions=[text], 138 | melody_wavs=melody, 139 | melody_sample_rate=msr, 140 | progress=False 141 | ) 142 | else: 143 | wav = MODEL.generate(descriptions=[text], progress=False) 144 | return wav 145 | 146 | 147 | def generate(model, text, melody, duration, topk, topp, temperature, cfg_coef, base_duration, 148 | sliding_window_seconds, continue_file, cf_cutoff, sc_text, seed, directory_name,finetuned_on): 149 | global MODEL 150 | if MODEL is None or MODEL.name != model: 151 | MODEL = load_model(model,directory_name,finetuned_on) 152 | 153 | final_length_seconds = duration 154 | descriptions = text 155 | 156 | topk = int(topk) 157 | int_seed = int(seed) 158 | cur_seed = set_seed(int_seed) 159 | print("seed: " + str(cur_seed)) 160 | 161 | if duration > 30: 162 | MODEL.set_generation_params( 163 | use_sampling=True, 164 | top_k=topk, 165 | top_p=topp, 166 | temperature=temperature, 167 | cfg_coef=cfg_coef, 168 | duration=base_duration, 169 | ) 170 | else: 171 | MODEL.set_generation_params( 172 | use_sampling=True, 173 | top_k=topk, 174 | top_p=topp, 175 | temperature=temperature, 176 | cfg_coef=cfg_coef, 177 | duration=duration, 178 | ) 179 | iterations_required = int(final_length_seconds / sliding_window_seconds) 180 | print(f"Iterations required: {iterations_required}") 181 | sr = MODEL.sample_rate 182 | print(f"Sample rate: {sr}") 183 | msr = None 184 | wav = None # wav shape will be [1, 1, sr * seconds] 185 | melody_boolean = False 186 | if melody: 187 | msr, melody = melody[0], torch.from_numpy(melody[1]).to(MODEL.device).float().t().unsqueeze(0) 188 | print(melody.shape) 189 | if melody.dim() == 2: 190 | melody = melody[None] 191 | melody = melody[..., :int(msr * MODEL.lm.cfg.dataset.segment_duration)] 192 | melody_boolean = True 193 | 194 | if (duration > 30): 195 | for i in range(iterations_required): 196 | print(f"Generating {i + 1}/{iterations_required}") 197 | if i == 0: 198 | wav = initial_generate(melody_boolean, MODEL, text, melody, msr, continue_file, base_duration, cf_cutoff, sc_text) 199 | wav = wav[:, :, :sr * sliding_window_seconds] 200 | else: 201 | new_chunk = None 202 | previous_chunk = wav[:, :, -sr * (base_duration - sliding_window_seconds):] 203 | if continue_file: 204 | if not sc_text: 205 | new_chunk = MODEL.generate_continuation(previous_chunk, prompt_sample_rate=sr, progress=False) 206 | else: 207 | new_chunk = MODEL.generate_continuation(previous_chunk, descriptions=[text], prompt_sample_rate=sr, progress=False) 208 | else: 209 | new_chunk = MODEL.generate_continuation(previous_chunk, descriptions=[text], prompt_sample_rate=sr, progress=False) 210 | wav = torch.cat((wav, new_chunk[:, :, -sr * sliding_window_seconds:]), dim=2) 211 | else: 212 | wav = initial_generate(melody_boolean, MODEL, text, melody, msr, continue_file, duration, cf_cutoff, sc_text) 213 | 214 | print(f"Final length: {wav.shape[2] / sr}s") 215 | output = wav.detach().cpu().float()[0] 216 | now = datetime.now() 217 | d = dirname(abspath(__file__)) 218 | file_name = d + "/results/" + now.strftime("%Y%m%d_%H%M%S") + "-" + str(cur_seed) + ".wav" 219 | with NamedTemporaryFile("wb", suffix=".wav", delete=False) as file: 220 | audio_write(file_name, output, MODEL.sample_rate, strategy="loudness", loudness_headroom_db=16, 221 | add_suffix=False, loudness_compressor=True) 222 | print(file_name) 223 | set_seed(-1) 224 | return file_name 225 | 226 | def get_datasets(path: str, ext: str): 227 | return ['None'] + glob(current_directory) 228 | 229 | def train_local(dataset_path: str, 230 | model_id: str, 231 | lr: float, 232 | epochs: int, 233 | use_wandb: bool, 234 | save_step: int = None,): 235 | if save_step==0: 236 | save_step=None 237 | wandb : int 238 | if use_wandb: 239 | wandb=1 240 | else: 241 | wandb=0 242 | train( 243 | dataset_path=dataset_path, 244 | model_id=model_id, 245 | lr=lr, 246 | epochs=int(epochs), 247 | use_wandb=wandb, 248 | save_step=save_step, 249 | ) 250 | 251 | with gr.Blocks(analytics_enabled=False) as demo: 252 | with gr.Tab("Inference"): 253 | gr.Markdown("""# MusicGen Inference""") 254 | with gr.Row(): 255 | with gr.Column(): 256 | with gr.Row(): 257 | text = gr.Text(label="Input Text", interactive=True) 258 | melody = gr.Audio(source="upload", type="numpy", label="Melody Condition (optional) SUPPORTS MELODY ONLY", interactive=True) 259 | continue_file = gr.Audio(source="upload", type="filepath", 260 | label="Song to continue (optional) SUPPORTS ALL MODELS", interactive=True) 261 | 262 | with gr.Row(): 263 | model = gr.Radio(["melody", "medium", "small", "large", "custom"], label="Model", value="small", interactive=True) 264 | directory_name= gr.Text(label="Finetuned DIRECTORY_NAME", interactive=True) 265 | finetuned_on = gr.Radio(["small", "medium", "large"], label="FINETUNED_ON model", value="small", interactive=True) 266 | with gr.Row(): 267 | duration = gr.Slider(minimum=1, maximum=300, value=30,step=1, label="Duration", interactive=True) 268 | base_duration = gr.Slider(minimum=1, maximum=30, value=30, step=1, label="Base duration", interactive=True) 269 | sliding_window_seconds = gr.Slider(minimum=1, maximum=30, value=15, step=1, label="Sliding window", interactive=True) 270 | cf_cutoff = gr.Slider(minimum=1, maximum=30, value=15, step=1, label="Continuing song cutoff", interactive=True) 271 | with gr.Row(): 272 | topk = gr.Number(label="Top-k", value=250, interactive=True) 273 | topp = gr.Number(label="Top-p", value=0, interactive=True) 274 | temperature = gr.Number(label="Temperature", value=1.0, interactive=True) 275 | cfg_coef = gr.Number(label="Classifier Free Guidance", value=3.0, interactive=True) 276 | with gr.Row(): 277 | sc_text = gr.Checkbox(label="Use text for song continuation.", value=True) 278 | seed = gr.Number(label="seed", value=-1, interactive=True) 279 | with gr.Row(): 280 | submit = gr.Button("Submit") 281 | with gr.Row(): 282 | output = gr.Audio(label="Generated Music", type="filepath") 283 | 284 | submit.click(generate, inputs=[model, text, melody, duration, topk, topp, temperature, 285 | cfg_coef, base_duration, sliding_window_seconds, continue_file, cf_cutoff, sc_text, seed,directory_name,finetuned_on], outputs=[output]) 286 | gr.Examples( 287 | fn=generate, 288 | examples=[ 289 | [ 290 | "An 80s driving pop song with heavy drums and synth pads in the background", 291 | "./repositories/audiocraft/assets/bach.mp3", 292 | "melody" 293 | ], 294 | [ 295 | "A cheerful country song with acoustic guitars", 296 | "./repositories/audiocraft/assets/bolero_ravel.mp3", 297 | "melody" 298 | ], 299 | [ 300 | "90s rock song with electric guitar and heavy drums", 301 | None, 302 | "medium" 303 | ], 304 | [ 305 | "a light and cheerly EDM track, with syncopated drums, aery pads, and strong emotions", 306 | "./repositories/audiocraft/assets/bach.mp3", 307 | "melody" 308 | ], 309 | [ 310 | "lofi slow bpm electro chill with organic samples", 311 | None, 312 | "medium", 313 | ], 314 | ], 315 | inputs=[text, melody, model], 316 | outputs=[output] 317 | ) 318 | gr.Markdown( 319 | """ 320 | This is a webui for MusicGen with 30+ second generation support. 321 | 322 | Models 323 | 1. Melody -- a music generation model capable of generating music condition on text and melody inputs. **Note**, you can also use text only. 324 | 2. Small -- a 300M transformer decoder conditioned on text only. 325 | 3. Medium -- a 1.5B transformer decoder conditioned on text only. 326 | 4. Large -- a 3.3B transformer decoder conditioned on text only (might OOM for the longest sequences.) - recommended for continuing songs 327 | 328 | When the optional melody conditioning wav is provided, the model will extract 329 | a broad melody and try to follow it in the generated samples. Only the first chunk of the song will 330 | be generated with melody conditioning, the others will just continue on the first chunk. 331 | 332 | Base duration of 30 seconds is recommended. 333 | 334 | Sliding window of 10/15/20 seconds is recommended. 335 | 336 | When continuing songs, a continuing song cutoff of 5 seconds gives good results. Continuing song cutoff - number of seconds to be taken from the end of the continuing song. 337 | 338 | Gradio analytics are disabled. 339 | """ 340 | ) 341 | with gr.Tab("Training"): 342 | with gr.Row(): 343 | with gr.Column(): 344 | dataset_path = gr.Dropdown(choices=glob.glob(current_directory+"/training/datasets/*/"), value='None', 345 | label='Dataset', info='The dataset path to use for training.', interactive=True) 346 | ui.create_refresh_button(dataset_path, lambda: None, 347 | lambda: {'choices': glob.glob(current_directory+"/training/datasets/*/")}, 348 | 'refresh-button') 349 | with gr.Column(): 350 | lr = gr.Number(label="Learning rate", value=0.0001, interactive=True) 351 | epochs = gr.Number(label="Epoch count", value=5, interactive=True) 352 | use_wandb = gr.Checkbox(label="Use WanDB", value=False, interactive=True) 353 | save_step = gr.Number(label="Number of steps after which to save a checkpoint. 0 is treated as none.", value=0, interactive=True) 354 | with gr.Row(): 355 | model_id = gr.Radio(["small", "medium", "large"], label="Model", value="small", interactive=True) 356 | train_button = gr.Button(label="Start training") 357 | train_button.click(train_local, inputs=[dataset_path,model_id, lr, epochs, use_wandb, save_step], outputs=[output]) 358 | gr.Markdown( 359 | """ 360 | # Training 361 | 362 | Model gets saved to models/ as `lm_final.pt` 363 | ### Using the finetuned model 364 | 365 | 1) Place it in models/DIRECTORY_NAME/ 366 | 2) In the Inference tab choose `custom` as the model and enter DIRECTORY_NAME into the input field. 367 | 3) In the Inference tab choose the model it was finetuned on 368 | 369 | ### Options 370 | 371 | - `dataset_path` path to your dataset with WAV and TXT pairs. 372 | - `model_id - MusicGen model to use. Can be `small`/`medium`/`large`. Default: `small` - model it will be finetuned on 373 | - `lr`: Float, learning rate. Default: `0.0001`/`1e-4` 374 | - `epochs`: Integer, epoch count. Default: `5` 375 | - `use_wandb`: Integer, `1` to enable wandb, `0` to disable it. Default: `0` = Disabled 376 | - `save_step`: Integer, amount of steps to save a checkpoint. Default: None 377 | 378 | Gradio analytics are disabled. 379 | """ 380 | ) 381 | 382 | if shared.args.listen: 383 | demo.launch(share=shared.args.share, server_name=shared.args.listen_host or '0.0.0.0', server_port=shared.args.listen_port, inbrowser=shared.args.auto_launch) 384 | else: 385 | demo.launch(share=shared.args.share, server_port=shared.args.listen_port, inbrowser=shared.args.auto_launch) 386 | -------------------------------------------------------------------------------- /webuibatch.py: -------------------------------------------------------------------------------- 1 | import contextlib 2 | from datetime import datetime 3 | import os 4 | import random 5 | import sys 6 | import typing 7 | import wave 8 | import glob 9 | from pathlib import Path 10 | from tempfile import NamedTemporaryFile 11 | 12 | import numpy as np 13 | import requests 14 | import torch 15 | import torchaudio 16 | from os.path import dirname, abspath 17 | from modules import shared, ui 18 | cuda = True 19 | #check for mps 20 | if torch.backends.mps.is_available(): 21 | mps_device = torch.device("mps") 22 | x = torch.ones(1, device=mps_device) 23 | cuda = False 24 | 25 | sys.path.insert(0, str(Path("repositories/audiocraft"))) 26 | sys.path.insert(0, str(Path("repositories/musicgen_trainer"))) 27 | from train import train 28 | from audiocraft.data.audio import audio_write 29 | from audiocraft.data.audio_utils import convert_audio 30 | from audiocraft.models import MusicGen 31 | 32 | os.environ['GRADIO_ANALYTICS_ENABLED'] = 'False' 33 | def my_get(url, **kwargs): 34 | kwargs.setdefault('allow_redirects', True) 35 | return requests.api.request('get', 'http://127.0.0.1/', **kwargs) 36 | original_get = requests.get 37 | requests.get = my_get 38 | import gradio as gr 39 | requests.get = original_get 40 | 41 | 42 | MODEL = None 43 | current_directory = dirname(abspath(__file__)) 44 | 45 | def load_model(version, DIRECTORY_NAME, FINETUNED_ON): 46 | if version != "custom": 47 | print("Loading model", version) 48 | path=current_directory+"/models/" + version + "/" 49 | if os.path.exists(path): 50 | model = MusicGen.get_pretrained(directory=path,name=version) 51 | 52 | else: model = MusicGen.get_pretrained(name=version) 53 | else: 54 | finetuned_dir =current_directory + "/models/" + DIRECTORY_NAME + "/" + "lm_final.pt" 55 | model= MusicGen.get_pretrained(name=FINETUNED_ON) 56 | model.lm.load_state_dict(torch.load(finetuned_dir)) 57 | model.name="custom" 58 | return model 59 | 60 | 61 | def set_seed(seed: int = 0): 62 | torch.backends.cudnn.deterministic = True 63 | torch.backends.cudnn.benchmark = False 64 | if seed <= 0: 65 | seed = np.random.default_rng().integers(1, 2**32 - 1) 66 | seed = np.uint32(seed).item() 67 | assert 0 < seed < 2**32 68 | original_seed = seed 69 | np.random.seed(seed) 70 | random.seed(seed) 71 | torch.manual_seed(seed) 72 | if cuda: 73 | torch.cuda.manual_seed_all(seed) 74 | os.environ["PYTHONHASHSEED"] = str(seed) 75 | return original_seed 76 | 77 | 78 | def generate_cmelody(descriptions: typing.List[str], melody_wavs: typing.Union[torch.Tensor, typing.List[typing.Optional[torch.Tensor]]], 79 | msr: int, prompt: torch.Tensor, psr: int, MODEL, progress: bool = False) -> torch.Tensor: 80 | if isinstance(melody_wavs, torch.Tensor): 81 | if melody_wavs.dim() == 2: 82 | melody_wavs = melody_wavs[None] 83 | if melody_wavs.dim() != 3: 84 | raise ValueError("melody_wavs should have a shape [B, C, T].") 85 | melody_wavs = list(melody_wavs) 86 | else: 87 | for melody in melody_wavs: 88 | if melody is not None: 89 | assert melody.dim() == 2, "one melody in the list has the wrong number of dims." 90 | 91 | melody_wavs = [ 92 | convert_audio(wav, msr, MODEL.sample_rate, MODEL.audio_channels) 93 | if wav is not None else None 94 | for wav in melody_wavs] 95 | 96 | if prompt.dim() == 2: 97 | prompt = prompt[None] 98 | if prompt.dim() != 3: 99 | raise ValueError("prompt should have 3 dimensions: [B, C, T] (C = 1).") 100 | prompt = convert_audio(prompt, psr, MODEL.sample_rate, MODEL.audio_channels) 101 | if descriptions is None: 102 | descriptions = [None] * len(prompt) 103 | attributes, prompt_tokens = MusicGen._prepare_tokens_and_attributes(MODEL, descriptions=descriptions, prompt=prompt, melody_wavs=melody_wavs) 104 | assert prompt_tokens is not None 105 | return MusicGen._generate_tokens(MODEL, attributes, prompt_tokens, progress) 106 | 107 | 108 | def initial_generate(melody_boolean, MODEL, text, melody, msr, continue_file, duration, cf_cutoff, sc_text): 109 | wav = None 110 | if continue_file: 111 | data_waveform, cfsr = (torchaudio.load(continue_file)) 112 | if cuda: 113 | wav = data_waveform.cuda() 114 | else: 115 | wav = data_waveform.mps_device() 116 | cf_len = 0 117 | with contextlib.closing(wave.open(continue_file, 'r')) as f: 118 | frames = f.getnframes() 119 | rate = f.getframerate() 120 | cf_len = frames / float(rate) 121 | if wav.dim() == 2: 122 | wav = wav[None] 123 | wav = wav[:, :, int(-cfsr * min(29, cf_len, duration - 1, cf_cutoff)):] 124 | new_chunk = None 125 | if not melody_boolean: 126 | if not sc_text: 127 | new_chunk = MODEL.generate_continuation(wav, prompt_sample_rate=cfsr, progress=False) 128 | else: 129 | new_chunk = MODEL.generate_continuation(wav, descriptions=[text], prompt_sample_rate=cfsr, progress=False) 130 | wav = new_chunk 131 | else: 132 | new_chunk = generate_cmelody([text], melody, msr, wav, cfsr, MODEL, progress=False) 133 | wav = new_chunk 134 | else: 135 | if melody_boolean: 136 | wav = MODEL.generate_with_chroma( 137 | descriptions=[text], 138 | melody_wavs=melody, 139 | melody_sample_rate=msr, 140 | progress=False 141 | ) 142 | else: 143 | wav = MODEL.generate(descriptions=[text], progress=False) 144 | return wav 145 | 146 | def generate_batch(model, text, melody, duration, topk, topp, temperature, cfg_coef, base_duration, 147 | sliding_window_seconds, continue_file, cf_cutoff, sc_text, seed, directory_name, 148 | finetuned_on, batch_mode, num_batches, infinite): 149 | num_batches = int(num_batches) # Convert num_batches to an integer 150 | if infinite: 151 | while True: 152 | generated_output = generate(model, text, melody, duration, topk, topp, temperature, cfg_coef, 153 | base_duration, sliding_window_seconds, continue_file, cf_cutoff, 154 | sc_text, seed, directory_name, finetuned_on) 155 | else: 156 | # Normal non-batch generation logic 157 | generated_output = generate(model, text, melody, duration, topk, topp, temperature, cfg_coef, 158 | base_duration, sliding_window_seconds, continue_file, cf_cutoff, 159 | sc_text, seed, directory_name, finetuned_on) 160 | output.value = generated_output 161 | 162 | def generate(model, text, melody, duration, topk, topp, temperature, cfg_coef, base_duration, 163 | sliding_window_seconds, continue_file, cf_cutoff, sc_text, seed, directory_name,finetuned_on): 164 | global MODEL 165 | if MODEL is None or MODEL.name != model: 166 | MODEL = load_model(model,directory_name,finetuned_on) 167 | 168 | final_length_seconds = duration 169 | descriptions = text 170 | 171 | topk = int(topk) 172 | int_seed = int(seed) 173 | cur_seed = set_seed(int_seed) 174 | print("seed: " + str(cur_seed)) 175 | 176 | if duration > 30: 177 | MODEL.set_generation_params( 178 | use_sampling=True, 179 | top_k=topk, 180 | top_p=topp, 181 | temperature=temperature, 182 | cfg_coef=cfg_coef, 183 | duration=base_duration, 184 | ) 185 | else: 186 | MODEL.set_generation_params( 187 | use_sampling=True, 188 | top_k=topk, 189 | top_p=topp, 190 | temperature=temperature, 191 | cfg_coef=cfg_coef, 192 | duration=duration, 193 | ) 194 | iterations_required = int(final_length_seconds / sliding_window_seconds) 195 | print(f"Iterations required: {iterations_required}") 196 | sr = MODEL.sample_rate 197 | print(f"Sample rate: {sr}") 198 | msr = None 199 | wav = None # wav shape will be [1, 1, sr * seconds] 200 | melody_boolean = False 201 | if melody: 202 | msr, melody = melody[0], torch.from_numpy(melody[1]).to(MODEL.device).float().t().unsqueeze(0) 203 | print(melody.shape) 204 | if melody.dim() == 2: 205 | melody = melody[None] 206 | melody = melody[..., :int(msr * MODEL.lm.cfg.dataset.segment_duration)] 207 | melody_boolean = True 208 | 209 | if (duration > 30): 210 | for i in range(iterations_required): 211 | print(f"Generating {i + 1}/{iterations_required}") 212 | if i == 0: 213 | wav = initial_generate(melody_boolean, MODEL, text, melody, msr, continue_file, base_duration, cf_cutoff, sc_text) 214 | wav = wav[:, :, :sr * sliding_window_seconds] 215 | else: 216 | new_chunk = None 217 | previous_chunk = wav[:, :, -sr * (base_duration - sliding_window_seconds):] 218 | if continue_file: 219 | if not sc_text: 220 | new_chunk = MODEL.generate_continuation(previous_chunk, prompt_sample_rate=sr, progress=False) 221 | else: 222 | new_chunk = MODEL.generate_continuation(previous_chunk, descriptions=[text], prompt_sample_rate=sr, progress=False) 223 | else: 224 | new_chunk = MODEL.generate_continuation(previous_chunk, descriptions=[text], prompt_sample_rate=sr, progress=False) 225 | wav = torch.cat((wav, new_chunk[:, :, -sr * sliding_window_seconds:]), dim=2) 226 | else: 227 | wav = initial_generate(melody_boolean, MODEL, text, melody, msr, continue_file, duration, cf_cutoff, sc_text) 228 | 229 | print(f"Final length: {wav.shape[2] / sr}s") 230 | output = wav.detach().cpu().float()[0] 231 | now = datetime.now() 232 | d = dirname(abspath(__file__)) 233 | file_name = d + "/results/" + now.strftime("%Y%m%d_%H%M%S") + "-" + str(cur_seed) + ".wav" 234 | with NamedTemporaryFile("wb", suffix=".wav", delete=False) as file: 235 | audio_write(file_name, output, MODEL.sample_rate, strategy="loudness", loudness_headroom_db=16, 236 | add_suffix=False, loudness_compressor=True) 237 | print(file_name) 238 | set_seed(-1) 239 | return file_name 240 | 241 | def get_datasets(path: str, ext: str): 242 | return ['None'] + glob(current_directory) 243 | 244 | def train_local(dataset_path: str, 245 | model_id: str, 246 | lr: float, 247 | epochs: int, 248 | use_wandb: bool, 249 | save_step: int = None,): 250 | if save_step==0: 251 | save_step=None 252 | wandb : int 253 | if use_wandb: 254 | wandb=1 255 | else: 256 | wandb=0 257 | train( 258 | dataset_path=dataset_path, 259 | model_id=model_id, 260 | lr=lr, 261 | epochs=int(epochs), 262 | use_wandb=wandb, 263 | save_step=save_step, 264 | ) 265 | 266 | with gr.Blocks(analytics_enabled=False) as demo: 267 | with gr.Tab("Inference"): 268 | gr.Markdown("""# MusicGen Inference""") 269 | with gr.Row(): 270 | with gr.Column(): 271 | with gr.Row(): 272 | batch_mode = gr.Checkbox(label="Batch Mode", value=False, interactive=True) 273 | num_batches = gr.Number(label="Number of Batches", value=5, interactive=True) 274 | infinite_checkbox = gr.Checkbox(label="Infinite", value=False, interactive=True) 275 | text = gr.Text(label="Input Text", interactive=True) 276 | melody = gr.Audio(source="upload", type="numpy", label="Melody Condition (optional) SUPPORTS MELODY ONLY", interactive=True) 277 | continue_file = gr.Audio(source="upload", type="filepath", 278 | label="Song to continue (optional) SUPPORTS ALL MODELS", interactive=True) 279 | 280 | with gr.Row(): 281 | model = gr.Radio(["melody", "medium", "small", "large", "custom"], label="Model", value="small", interactive=True) 282 | directory_name= gr.Text(label="Finetuned DIRECTORY_NAME", interactive=True) 283 | finetuned_on = gr.Radio(["small", "medium", "large"], label="FINETUNED_ON model", value="small", interactive=True) 284 | with gr.Row(): 285 | duration = gr.Slider(minimum=1, maximum=300, value=30,step=1, label="Duration", interactive=True) 286 | base_duration = gr.Slider(minimum=1, maximum=30, value=30, step=1, label="Base duration", interactive=True) 287 | sliding_window_seconds = gr.Slider(minimum=1, maximum=30, value=15, step=1, label="Sliding window", interactive=True) 288 | cf_cutoff = gr.Slider(minimum=1, maximum=30, value=15, step=1, label="Continuing song cutoff", interactive=True) 289 | with gr.Row(): 290 | topk = gr.Number(label="Top-k", value=250, interactive=True) 291 | topp = gr.Number(label="Top-p", value=0, interactive=True) 292 | temperature = gr.Number(label="Temperature", value=1.0, interactive=True) 293 | cfg_coef = gr.Number(label="Classifier Free Guidance", value=3.0, interactive=True) 294 | with gr.Row(): 295 | sc_text = gr.Checkbox(label="Use text for song continuation.", value=True) 296 | seed = gr.Number(label="seed", value=-1, interactive=True) 297 | with gr.Row(): 298 | submit = gr.Button("Submit") 299 | with gr.Row(): 300 | output = gr.Audio(label="Generated Music", type="filepath") 301 | 302 | submit.click(generate_batch, inputs=[model, text, melody, duration, topk, topp, temperature, 303 | cfg_coef, base_duration, sliding_window_seconds, continue_file, 304 | cf_cutoff, sc_text, seed, directory_name, finetuned_on, 305 | batch_mode, num_batches, infinite_checkbox], 306 | outputs=[output]) 307 | gr.Examples( 308 | fn=generate, 309 | examples=[ 310 | [ 311 | "An 80s driving pop song with heavy drums and synth pads in the background", 312 | "./repositories/audiocraft/assets/bach.mp3", 313 | "melody" 314 | ], 315 | [ 316 | "A cheerful country song with acoustic guitars", 317 | "./repositories/audiocraft/assets/bolero_ravel.mp3", 318 | "melody" 319 | ], 320 | [ 321 | "90s rock song with electric guitar and heavy drums", 322 | None, 323 | "medium" 324 | ], 325 | [ 326 | "a light and cheerly EDM track, with syncopated drums, aery pads, and strong emotions", 327 | "./repositories/audiocraft/assets/bach.mp3", 328 | "melody" 329 | ], 330 | [ 331 | "lofi slow bpm electro chill with organic samples", 332 | None, 333 | "medium", 334 | ], 335 | ], 336 | inputs=[text, melody, model], 337 | outputs=[output] 338 | ) 339 | gr.Markdown( 340 | """ 341 | This is a webui for MusicGen with 30+ second generation support. 342 | 343 | Models 344 | 1. Melody -- a music generation model capable of generating music condition on text and melody inputs. **Note**, you can also use text only. 345 | 2. Small -- a 300M transformer decoder conditioned on text only. 346 | 3. Medium -- a 1.5B transformer decoder conditioned on text only. 347 | 4. Large -- a 3.3B transformer decoder conditioned on text only (might OOM for the longest sequences.) - recommended for continuing songs 348 | 349 | When the optional melody conditioning wav is provided, the model will extract 350 | a broad melody and try to follow it in the generated samples. Only the first chunk of the song will 351 | be generated with melody conditioning, the others will just continue on the first chunk. 352 | 353 | Base duration of 30 seconds is recommended. 354 | 355 | Sliding window of 10/15/20 seconds is recommended. 356 | 357 | When continuing songs, a continuing song cutoff of 5 seconds gives good results. Continuing song cutoff - number of seconds to be taken from the end of the continuing song. 358 | 359 | Gradio analytics are disabled. 360 | """ 361 | ) 362 | with gr.Tab("Training"): 363 | with gr.Row(): 364 | with gr.Column(): 365 | dataset_path = gr.Dropdown(choices=glob.glob(current_directory+"/training/datasets/*/"), value='None', 366 | label='Dataset', info='The dataset path to use for training.', interactive=True) 367 | ui.create_refresh_button(dataset_path, lambda: None, 368 | lambda: {'choices': glob.glob(current_directory+"/training/datasets/*/")}, 369 | 'refresh-button') 370 | with gr.Column(): 371 | lr = gr.Number(label="Learning rate", value=0.0001, interactive=True) 372 | epochs = gr.Number(label="Epoch count", value=5, interactive=True) 373 | use_wandb = gr.Checkbox(label="Use WanDB", value=False, interactive=True) 374 | save_step = gr.Number(label="Number of steps after which to save a checkpoint. 0 is treated as none.", value=0, interactive=True) 375 | with gr.Row(): 376 | model_id = gr.Radio(["small", "medium", "large"], label="Model", value="small", interactive=True) 377 | train_button = gr.Button(label="Start training") 378 | train_button.click(train_local, inputs=[dataset_path,model_id, lr, epochs, use_wandb, save_step], outputs=[output]) 379 | gr.Markdown( 380 | """ 381 | # Training 382 | 383 | Model gets saved to models/ as `lm_final.pt` 384 | ### Using the finetuned model 385 | 386 | 1) Place it in models/DIRECTORY_NAME/ 387 | 2) In the Inference tab choose `custom` as the model and enter DIRECTORY_NAME into the input field. 388 | 3) In the Inference tab choose the model it was finetuned on 389 | 390 | ### Options 391 | 392 | - `dataset_path` path to your dataset with WAV and TXT pairs. 393 | - `model_id - MusicGen model to use. Can be `small`/`medium`/`large`. Default: `small` - model it will be finetuned on 394 | - `lr`: Float, learning rate. Default: `0.0001`/`1e-4` 395 | - `epochs`: Integer, epoch count. Default: `5` 396 | - `use_wandb`: Integer, `1` to enable wandb, `0` to disable it. Default: `0` = Disabled 397 | - `save_step`: Integer, amount of steps to save a checkpoint. Default: None 398 | 399 | Gradio analytics are disabled. 400 | """ 401 | ) 402 | 403 | if shared.args.listen: 404 | demo.launch(share=shared.args.share, server_name=shared.args.listen_host or '0.0.0.0', server_port=shared.args.listen_port, inbrowser=shared.args.auto_launch) 405 | else: 406 | demo.launch(share=shared.args.share, server_port=shared.args.listen_port, inbrowser=shared.args.auto_launch) 407 | --------------------------------------------------------------------------------