Upload 13 files

.gitignore

@@ -0,0 +1,2 @@
+__pycache__
+.venv

LICENSE.md

@@ -0,0 +1,660 @@
+### GNU AFFERO GENERAL PUBLIC LICENSE
+
+Version 3, 19 November 2007
+
+Copyright (C) 2007 Free Software Foundation, Inc.
+<https://fsf.org/>
+
+Everyone is permitted to copy and distribute verbatim copies of this
+license document, but changing it is not allowed.
+
+### Preamble
+
+The GNU Affero General Public License is a free, copyleft license for
+software and other kinds of works, specifically designed to ensure
+cooperation with the community in the case of network server software.
+
+The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works. By contrast,
+our General Public Licenses are intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains
+free software for all its users.
+
+When we speak of free software, we are referring to freedom, not
+price. Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+Developers that use our General Public Licenses protect your rights
+with two steps: (1) assert copyright on the software, and (2) offer
+you this License which gives you legal permission to copy, distribute
+and/or modify the software.
+
+A secondary benefit of defending all users' freedom is that
+improvements made in alternate versions of the program, if they
+receive widespread use, become available for other developers to
+incorporate. Many developers of free software are heartened and
+encouraged by the resulting cooperation. However, in the case of
+software used on network servers, this result may fail to come about.
+The GNU General Public License permits making a modified version and
+letting the public access it on a server without ever releasing its
+source code to the public.
+
+The GNU Affero General Public License is designed specifically to
+ensure that, in such cases, the modified source code becomes available
+to the community. It requires the operator of a network server to
+provide the source code of the modified version running there to the
+users of that server. Therefore, public use of a modified version, on
+a publicly accessible server, gives the public access to the source
+code of the modified version.
+
+An older license, called the Affero General Public License and
+published by Affero, was designed to accomplish similar goals. This is
+a different license, not a version of the Affero GPL, but Affero has
+released a new version of the Affero GPL which permits relicensing
+under this license.
+
+The precise terms and conditions for copying, distribution and
+modification follow.
+
+### TERMS AND CONDITIONS
+
+#### 0. Definitions.
+
+"This License" refers to version 3 of the GNU Affero General Public
+License.
+
+"Copyright" also means copyright-like laws that apply to other kinds
+of works, such as semiconductor masks.
+
+"The Program" refers to any copyrightable work licensed under this
+License. Each licensee is addressed as "you". "Licensees" and
+"recipients" may be individuals or organizations.
+
+To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of
+an exact copy. The resulting work is called a "modified version" of
+the earlier work or a work "based on" the earlier work.
+
+A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy. Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies. Mere interaction with a user
+through a computer network, with no transfer of a copy, is not
+conveying.
+
+An interactive user interface displays "Appropriate Legal Notices" to
+the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License. If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+#### 1. Source Code.
+
+The "source code" for a work means the preferred form of the work for
+making modifications to it. "Object code" means any non-source form of
+a work.
+
+A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form. A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities. However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work. For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+The Corresponding Source need not include anything that users can
+regenerate automatically from other parts of the Corresponding Source.
+
+The Corresponding Source for a work in source code form is that same
+work.
+
+#### 2. Basic Permissions.
+
+All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met. This License explicitly affirms your unlimited
+permission to run the unmodified Program. The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work. This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+You may make, run and propagate covered works that you do not convey,
+without conditions so long as your license otherwise remains in force.
+You may convey covered works to others for the sole purpose of having
+them make modifications exclusively for you, or provide you with
+facilities for running those works, provided that you comply with the
+terms of this License in conveying all material for which you do not
+control copyright. Those thus making or running the covered works for
+you must do so exclusively on your behalf, under your direction and
+control, on terms that prohibit them from making any copies of your
+copyrighted material outside their relationship with you.
+
+Conveying under any other circumstances is permitted solely under the
+conditions stated below. Sublicensing is not allowed; section 10 makes
+it unnecessary.
+
+#### 3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such
+circumvention is effected by exercising rights under this License with
+respect to the covered work, and you disclaim any intention to limit
+operation or modification of the work as a means of enforcing, against
+the work's users, your or third parties' legal rights to forbid
+circumvention of technological measures.
+
+#### 4. Conveying Verbatim Copies.
+
+You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+#### 5. Conveying Modified Source Versions.
+
+You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these
+conditions:
+
+-   a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+-   b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under
+    section 7. This requirement modifies the requirement in section 4
+    to "keep intact all notices".
+-   c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy. This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged. This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+-   d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit. Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+#### 6. Conveying Non-Source Forms.
+
+You may convey a covered work in object code form under the terms of
+sections 4 and 5, provided that you also convey the machine-readable
+Corresponding Source under the terms of this License, in one of these
+ways:
+
+-   a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+-   b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the Corresponding
+    Source from a network server at no charge.
+-   c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source. This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+-   d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge. You need not require recipients to copy the
+    Corresponding Source along with the object code. If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source. Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+-   e) Convey the object code using peer-to-peer transmission,
+    provided you inform other peers where the object code and
+    Corresponding Source of the work are being offered to the general
+    public at no charge under subsection 6d.
+
+A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal,
+family, or household purposes, or (2) anything designed or sold for
+incorporation into a dwelling. In determining whether a product is a
+consumer product, doubtful cases shall be resolved in favor of
+coverage. For a particular product received by a particular user,
+"normally used" refers to a typical or common use of that class of
+product, regardless of the status of the particular user or of the way
+in which the particular user actually uses, or expects or is expected
+to use, the product. A product is a consumer product regardless of
+whether the product has substantial commercial, industrial or
+non-consumer uses, unless such uses represent the only significant
+mode of use of the product.
+
+"Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to
+install and execute modified versions of a covered work in that User
+Product from a modified version of its Corresponding Source. The
+information must suffice to ensure that the continued functioning of
+the modified object code is in no case prevented or interfered with
+solely because modification has been made.
+
+If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information. But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or
+updates for a work that has been modified or installed by the
+recipient, or for the User Product in which it has been modified or
+installed. Access to a network may be denied when the modification
+itself materially and adversely affects the operation of the network
+or violates the rules and protocols for communication across the
+network.
+
+Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+#### 7. Additional Terms.
+
+"Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law. If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it. (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.) You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders
+of that material) supplement the terms of this License with terms:
+
+-   a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+-   b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+-   c) Prohibiting misrepresentation of the origin of that material,
+    or requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+-   d) Limiting the use for publicity purposes of names of licensors
+    or authors of the material; or
+-   e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+-   f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions
+    of it) with contractual assumptions of liability to the recipient,
+    for any liability that these contractual assumptions directly
+    impose on those licensors and authors.
+
+All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10. If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term. If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions; the
+above requirements apply either way.
+
+#### 8. Termination.
+
+You may not propagate or modify a covered work except as expressly
+provided under this License. Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+However, if you cease all violation of this License, then your license
+from a particular copyright holder is reinstated (a) provisionally,
+unless and until the copyright holder explicitly and finally
+terminates your license, and (b) permanently, if the copyright holder
+fails to notify you of the violation by some reasonable means prior to
+60 days after the cessation.
+
+Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License. If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+#### 9. Acceptance Not Required for Having Copies.
+
+You are not required to accept this License in order to receive or run
+a copy of the Program. Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance. However,
+nothing other than this License grants you permission to propagate or
+modify any covered work. These actions infringe copyright if you do
+not accept this License. Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+#### 10. Automatic Licensing of Downstream Recipients.
+
+Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License. You are not responsible
+for enforcing compliance by third parties with this License.
+
+An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations. If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License. For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+#### 11. Patents.
+
+A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based. The
+work thus licensed is called the contributor's "contributor version".
+
+A contributor's "essential patent claims" are all patent claims owned
+or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version. For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement). To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients. "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+A patent license is "discriminatory" if it does not include within the
+scope of its coverage, prohibits the exercise of, or is conditioned on
+the non-exercise of one or more of the rights that are specifically
+granted under this License. You may not convey a covered work if you
+are a party to an arrangement with a third party that is in the
+business of distributing software, under which you make payment to the
+third party based on the extent of your activity of conveying the
+work, and under which the third party grants, to any of the parties
+who would receive the covered work from you, a discriminatory patent
+license (a) in connection with copies of the covered work conveyed by
+you (or copies made from those copies), or (b) primarily for and in
+connection with specific products or compilations that contain the
+covered work, unless you entered into that arrangement, or that patent
+license was granted, prior to 28 March 2007.
+
+Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+#### 12. No Surrender of Others' Freedom.
+
+If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License. If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under
+this License and any other pertinent obligations, then as a
+consequence you may not convey it at all. For example, if you agree to
+terms that obligate you to collect a royalty for further conveying
+from those to whom you convey the Program, the only way you could
+satisfy both those terms and this License would be to refrain entirely
+from conveying the Program.
+
+#### 13. Remote Network Interaction; Use with the GNU General Public License.
+
+Notwithstanding any other provision of this License, if you modify the
+Program, your modified version must prominently offer all users
+interacting with it remotely through a computer network (if your
+version supports such interaction) an opportunity to receive the
+Corresponding Source of your version by providing access to the
+Corresponding Source from a network server at no charge, through some
+standard or customary means of facilitating copying of software. This
+Corresponding Source shall include the Corresponding Source for any
+work covered by version 3 of the GNU General Public License that is
+incorporated pursuant to the following paragraph.
+
+Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU General Public License into a single
+combined work, and to convey the resulting work. The terms of this
+License will continue to apply to the part which is the covered work,
+but the work with which it is combined will remain governed by version
+3 of the GNU General Public License.
+
+#### 14. Revised Versions of this License.
+
+The Free Software Foundation may publish revised and/or new versions
+of the GNU Affero General Public License from time to time. Such new
+versions will be similar in spirit to the present version, but may
+differ in detail to address new problems or concerns.
+
+Each version is given a distinguishing version number. If the Program
+specifies that a certain numbered version of the GNU Affero General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation. If the Program does not specify a version number of the
+GNU Affero General Public License, you may choose any version ever
+published by the Free Software Foundation.
+
+If the Program specifies that a proxy can decide which future versions
+of the GNU Affero General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+Later license versions may give you additional or different
+permissions. However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+#### 15. Disclaimer of Warranty.
+
+THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT
+WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND
+PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE
+DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR
+CORRECTION.
+
+#### 16. Limitation of Liability.
+
+IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR
+CONVEYS THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES,
+INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES
+ARISING OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT
+NOT LIMITED TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR
+LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM
+TO OPERATE WITH ANY OTHER PROGRAMS), EVEN IF SUCH HOLDER OR OTHER
+PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
+
+#### 17. Interpretation of Sections 15 and 16.
+
+If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+END OF TERMS AND CONDITIONS
+
+### How to Apply These Terms to Your New Programs
+
+If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these
+terms.
+
+To do so, attach the following notices to the program. It is safest to
+attach them to the start of each source file to most effectively state
+the exclusion of warranty; and each file should have at least the
+"copyright" line and a pointer to where the full notice is found.
+
+        <one line to give the program's name and a brief idea of what it does.>
+        Copyright (C) <year>  <name of author>
+
+        This program is free software: you can redistribute it and/or modify
+        it under the terms of the GNU Affero General Public License as
+        published by the Free Software Foundation, either version 3 of the
+        License, or (at your option) any later version.
+
+        This program is distributed in the hope that it will be useful,
+        but WITHOUT ANY WARRANTY; without even the implied warranty of
+        MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+        GNU Affero General Public License for more details.
+
+        You should have received a copy of the GNU Affero General Public License
+        along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper
+mail.
+
+If your software can interact with users remotely through a computer
+network, you should also make sure that it provides a way for users to
+get its source. For example, if your program is a web application, its
+interface could display a "Source" link that leads users to an archive
+of the code. There are many ways you could offer source, and different
+solutions will be better for different programs; see section 13 for
+the specific requirements.
+
+You should also get your employer (if you work as a programmer) or
+school, if any, to sign a "copyright disclaimer" for the program, if
+necessary. For more information on this, and how to apply and follow
+the GNU AGPL, see <https://www.gnu.org/licenses/>.

README.md

@@ -1,12 +1,5 @@
----
-title: Gradiopyg
-emoji: 🚀
-colorFrom: pink
-colorTo: gray
-sdk: gradio
-sdk_version: 3.29.0
-app_file: app.py
-pinned: false
----
+# gradio-ui
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+This is a prototype Gradio-based UI which allows you to chat with the [Pygmalion models](https://huggingface.co/PygmalionAI).
+
+We provide a notebook which directly interacts with this code in our repo, which can be found [here](./notebooks/GPU.ipynb). The notebook can either be run locally or can be used on [Google Colab](https://colab.research.google.com/github/PygmalionAI/gradio-ui/blob/master/notebooks/GPU.ipynb).

notebooks/GPU.ipynb

@@ -0,0 +1,46 @@
+{
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "metadata": {
+        "colab_type": "text",
+        "id": "view-in-github"
+      },
+      "source": [
+        "<a href=\"https://colab.research.google.com/github/PygmalionAI/gradio-ui/blob/master/notebooks/GPU.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
+      ]
+    },
+    {
+      "attachments": {},
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": [
+        "Due to Colab cracking down on this notebook, we've been forced to take it offline for a while. We're really, really sorry about this.\n",
+        "Please find other, alternative ways to use Pygmalion 6B.\n",
+        "\n",
+        "We'll try to be back soon. Our deepest apologies."
+      ]
+    },
+    {
+      "cell_type": "markdown",
+      "metadata": {},
+      "source": []
+    }
+  ],
+  "metadata": {
+    "accelerator": "GPU",
+    "colab": {
+      "provenance": []
+    },
+    "gpuClass": "standard",
+    "kernelspec": {
+      "display_name": "Python 3",
+      "name": "python3"
+    },
+    "language_info": {
+      "name": "python"
+    }
+  },
+  "nbformat": 4,
+  "nbformat_minor": 0
+}

requirements-local.txt

@@ -0,0 +1,3 @@
+accelerate
+safetensors
+transformers

requirements.txt

@@ -0,0 +1,2 @@
+gradio<=3.18.0
+requests

src/app.py

@@ -0,0 +1,131 @@
+#!/usr/bin/env python3
+import argparse
+import logging
+import typing as t
+
+from gradio_ui import build_gradio_ui_for
+from koboldai_client import run_raw_inference_on_kai, wait_for_kai_server
+from parsing import parse_messages_from_str
+from prompting import build_prompt_for
+from utils import clear_stdout
+
+logger = logging.getLogger(__name__)
+logging.basicConfig(level=logging.INFO)
+
+# For UI debugging purposes.
+DONT_USE_MODEL = False
+
+
+def main(server_port: int,
+         share_gradio_link: bool = False,
+         model_name: t.Optional[str] = None,
+         koboldai_url: t.Optional[str] = None) -> None:
+    '''Script entrypoint.'''
+    if model_name and not DONT_USE_MODEL:
+        from model import build_model_and_tokenizer_for, run_raw_inference
+        model, tokenizer = build_model_and_tokenizer_for(model_name)
+    else:
+        model, tokenizer = None, None
+
+    def inference_fn(history: t.List[str], user_input: str,
+                     generation_settings: t.Dict[str, t.Any],
+                     *char_settings: t.Any) -> str:
+        if DONT_USE_MODEL:
+            return "Mock response for UI tests."
+
+        # Brittle. Comes from the order defined in gradio_ui.py.
+        [
+            char_name,
+            _user_name,
+            char_persona,
+            char_greeting,
+            world_scenario,
+            example_dialogue,
+        ] = char_settings
+
+        # If we're just starting the conversation and the character has a greeting
+        # configured, return that instead. This is a workaround for the fact that
+        # Gradio assumed that a chatbot cannot possibly start a conversation, so we
+        # can't just have the greeting there automatically, it needs to be in
+        # response to a user message.
+        if len(history) == 0 and char_greeting is not None:
+            return f"{char_name}: {char_greeting}"
+
+        prompt = build_prompt_for(history=history,
+                                  user_message=user_input,
+                                  char_name=char_name,
+                                  char_persona=char_persona,
+                                  example_dialogue=example_dialogue,
+                                  world_scenario=world_scenario)
+
+        if model and tokenizer:
+            model_output = run_raw_inference(model, tokenizer, prompt,
+                                             user_input, **generation_settings)
+        elif koboldai_url:
+            model_output = f"{char_name}:"
+            model_output += run_raw_inference_on_kai(koboldai_url, prompt,
+                                                     **generation_settings)
+        else:
+            raise Exception(
+                "Not using local inference, but no Kobold instance URL was"
+                " given. Nowhere to perform inference on.")
+
+        generated_messages = parse_messages_from_str(model_output,
+                                                     ["You", char_name])
+        logger.debug("Parsed model response is: `%s`", generated_messages)
+        bot_message = generated_messages[0]
+
+        return bot_message
+
+    ui = build_gradio_ui_for(inference_fn, for_kobold=koboldai_url is not None)
+    ui.launch(server_port=server_port, share=share_gradio_link)
+
+
+def _parse_args_from_argv() -> argparse.Namespace:
+    '''Parses arguments coming in from the command line.'''
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-m",
+        "--model-name",
+        help=
+        "HuggingFace Transformers model name, if not using a KoboldAI instance as an inference server.",
+    )
+    parser.add_argument(
+        "-p",
+        "--port",
+        type=int,
+        default=3000,
+        help="Port to listen on.",
+    )
+    parser.add_argument(
+        "-k",
+        "--koboldai-url",
+        help="URL to a KoboldAI instance to use as an inference server.",
+    )
+    parser.add_argument(
+        "-s",
+        "--share",
+        action="store_true",
+        help="Enable to generate a public link for the Gradio UI.",
+    )
+
+    return parser.parse_args()
+
+
+if __name__ == "__main__":
+    args = _parse_args_from_argv()
+
+    if args.koboldai_url:
+        # I have no idea how long a safe wait time is, but we'd rather wait for
+        # too long rather than just cut the user off _right_ when the setup is
+        # about to finish, so let's pick something absurd here.
+        wait_for_kai_server(args.koboldai_url, max_wait_time_seconds=60 * 30)
+
+        # Clear out any Kobold logs so the user can clearly see the Gradio link
+        # that's about to show up afterwards.
+        clear_stdout()
+
+    main(model_name=args.model_name,
+         server_port=args.port,
+         koboldai_url=args.koboldai_url,
+         share_gradio_link=args.share)

src/gradio_ui.py

@@ -0,0 +1,442 @@
+import json
+import logging
+
+import gradio as gr
+
+
+def get_generation_defaults(for_kobold):
+    defaults = {
+        "do_sample": True,
+        "max_new_tokens": 196,
+        "temperature": 0.5,
+        "top_p": 0.9,
+        "top_k": 0,
+        "typical_p": 1.0,
+        "repetition_penalty": 1.05,
+    }
+
+    if for_kobold:
+        defaults.update({"max_context_length": 768})
+    else:
+        defaults.update({"penalty_alpha": 0.6})
+
+    return defaults
+
+
+logger = logging.getLogger(__name__)
+
+
+def build_gradio_ui_for(inference_fn, for_kobold):
+    '''
+    Builds a Gradio UI to interact with the model. Big thanks to TearGosling for
+    the initial version that inspired this.
+    '''
+    with gr.Blocks(title="Pygmalion", analytics_enabled=False) as interface:
+        history_for_gradio = gr.State([])
+        history_for_model = gr.State([])
+        generation_settings = gr.State(
+            get_generation_defaults(for_kobold=for_kobold))
+
+        def _update_generation_settings(
+            original_settings,
+            param_name,
+            new_value,
+        ):
+            '''
+            Merges `{param_name: new_value}` into `original_settings` and
+            returns a new dictionary.
+            '''
+            updated_settings = {**original_settings, param_name: new_value}
+            logging.debug("Generation settings updated to: `%s`",
+                         updated_settings)
+            return updated_settings
+
+        def _run_inference(
+            model_history,
+            gradio_history,
+            user_input,
+            generation_settings,
+            *char_setting_states,
+        ):
+            '''
+            Runs inference on the model, and formats the returned response for
+            the Gradio state and chatbot component.
+            '''
+            char_name = char_setting_states[0]
+            user_name = char_setting_states[1]
+
+            # If user input is blank, format it as if user was silent
+            if user_input is None or user_input.strip() == "":
+                user_input = "..."
+
+            inference_result = inference_fn(model_history, user_input,
+                                            generation_settings,
+                                            *char_setting_states)
+
+            inference_result_for_gradio = inference_result \
+                .replace(f"{char_name}:", f"**{char_name}:**") \
+                .replace("<USER>", user_name) \
+                .replace("\n", "<br>") # Gradio chatbot component can display br tag as linebreak
+
+            model_history.append(f"You: {user_input}")
+            model_history.append(inference_result)
+            gradio_history.append((user_input, inference_result_for_gradio))
+
+            return None, model_history, gradio_history, gradio_history
+
+        def _regenerate(
+            model_history,
+            gradio_history,
+            generation_settings,
+            *char_setting_states,
+        ):
+            '''Regenerates the last response.'''
+            return _run_inference(
+                model_history[:-2],
+                gradio_history[:-1],
+                model_history[-2].replace("You: ", ""),
+                generation_settings,
+                *char_setting_states,
+            )
+
+        def _undo_last_exchange(model_history, gradio_history):
+            '''Undoes the last exchange (message pair).'''
+            return model_history[:-2], gradio_history[:-1], gradio_history[:-1]
+
+        def _save_chat_history(model_history, *char_setting_states):
+            '''Saves the current chat history to a .json file.'''
+            char_name = char_setting_states[0]
+            with open(f"{char_name}_conversation.json", "w") as f:
+                f.write(json.dumps({"chat": model_history}))
+            return f"{char_name}_conversation.json"
+
+        def _load_chat_history(file_obj, *char_setting_states):
+            '''Loads up a chat history from a .json file.'''
+            # #############################################################################################
+            # TODO(TG): Automatically detect and convert any CAI dump files loaded in to Pygmalion format #
+            # #############################################################################################
+
+            # https://stackoverflow.com/questions/5389507/iterating-over-every-two-elements-in-a-list
+            def pairwise(iterable):
+                # "s -> (s0, s1), (s2, s3), (s4, s5), ..."
+                a = iter(iterable)
+                return zip(a, a)
+
+            char_name = char_setting_states[0]
+            user_name = char_setting_states[1]
+
+            file_data = json.loads(file_obj.decode('utf-8'))
+            model_history = file_data["chat"]
+            # Construct a new gradio history
+            new_gradio_history = []
+            for human_turn, bot_turn in pairwise(model_history):
+                # Handle the situation where convo history may be loaded before character defs
+                if char_name == "":
+                    # Grab char name from the model history
+                    char_name = bot_turn.split(":")[0]
+                # Format the user and bot utterances
+                user_turn = human_turn.replace("You: ", "")
+                bot_turn = bot_turn.replace(f"{char_name}:", f"**{char_name}:**")
+
+                # Somebody released a script on /g/ which tries to convert CAI dump logs
+                # to Pygmalion character settings and chats. The anonymization of the dumps, however, means that
+                # [NAME_IN_MESSAGE_REDACTED] is left in the conversational history. We obviously wouldn't want this
+                # This therefore accomodates users of that script, so that [NAME_IN_MESSAGE_REDACTED] doesn't have
+                # to be manually edited in the conversation JSON.
+                # The model shouldn't generate [NAME_IN_MESSAGE_REDACTED] by itself.
+                user_turn = user_turn.replace("[NAME_IN_MESSAGE_REDACTED]", user_name)
+                bot_turn = bot_turn.replace("[NAME_IN_MESSAGE_REDACTED]", user_name)
+
+                new_gradio_history.append((user_turn, bot_turn))
+
+            return model_history, new_gradio_history, new_gradio_history
+
+        with gr.Tab("Character Settings") as settings_tab:
+            charfile, char_setting_states = _build_character_settings_ui()
+
+        with gr.Tab("Chat Window"):
+            chatbot = gr.Chatbot(
+                label="Your conversation will show up here").style(
+                    color_map=("#326efd", "#212528"))
+
+            char_name, _user_name, char_persona, char_greeting, world_scenario, example_dialogue = char_setting_states
+            charfile.upload(
+                fn=_char_file_upload,
+                inputs=[charfile, history_for_model, history_for_gradio],
+                outputs=[history_for_model, history_for_gradio, chatbot, char_name, char_persona, char_greeting, world_scenario, example_dialogue]
+            )
+
+            message = gr.Textbox(
+                label="Your message (hit Enter to send)",
+                placeholder="Write a message...",
+            )
+            message.submit(
+                fn=_run_inference,
+                inputs=[
+                    history_for_model, history_for_gradio, message,
+                    generation_settings, *char_setting_states
+                ],
+                outputs=[
+                    message, history_for_model, history_for_gradio, chatbot
+                ],
+            )
+
+            with gr.Row():
+                send_btn = gr.Button("Send", variant="primary")
+                send_btn.click(
+                    fn=_run_inference,
+                    inputs=[
+                        history_for_model, history_for_gradio, message,
+                        generation_settings, *char_setting_states
+                    ],
+                    outputs=[
+                        message, history_for_model, history_for_gradio, chatbot
+                    ],
+                )
+
+                regenerate_btn = gr.Button("Regenerate")
+                regenerate_btn.click(
+                    fn=_regenerate,
+                    inputs=[
+                        history_for_model, history_for_gradio,
+                        generation_settings, *char_setting_states
+                    ],
+                    outputs=[
+                        message, history_for_model, history_for_gradio, chatbot
+                    ],
+                )
+
+                undo_btn = gr.Button("Undo last exchange")
+                undo_btn.click(
+                    fn=_undo_last_exchange,
+                    inputs=[history_for_model, history_for_gradio],
+                    outputs=[history_for_model, history_for_gradio, chatbot],
+                )
+
+            with gr.Row():
+                with gr.Column():
+                    chatfile = gr.File(type="binary", file_types=[".json"], interactive=True)
+                    chatfile.upload(
+                        fn=_load_chat_history,
+                        inputs=[chatfile, *char_setting_states],
+                        outputs=[history_for_model, history_for_gradio, chatbot]
+                    )
+
+                    save_char_btn = gr.Button(value="Save Conversation History")
+                    save_char_btn.click(_save_chat_history, inputs=[history_for_model, *char_setting_states], outputs=[chatfile])
+                with gr.Column():
+                    gr.Markdown("""
+                        ### To save a chat
+                        Click "Save Conversation History". The file will appear above the button and you can click to download it.
+
+                        ### To load a chat
+                        Drag a valid .json file onto the upload box, or click the box to browse.
+
+                        **Remember to fill out/load up your character definitions before resuming a chat!**
+                    """)
+
+
+
+        with gr.Tab("Generation Settings"):
+            _build_generation_settings_ui(
+                state=generation_settings,
+                fn=_update_generation_settings,
+                for_kobold=for_kobold,
+            )
+
+    return interface
+
+
+def _char_file_upload(file_obj, history_model, history_gradio):
+    file_data = json.loads(file_obj.decode('utf-8'))
+    char_name = file_data["char_name"]
+    greeting = file_data["char_greeting"]
+    empty_history = not history_model or (len(history_model) <= 2 and history_model[0] == '')
+    if empty_history and char_name and greeting:
+        # if chat history is empty so far, and there is a character greeting, add character greeting to the chat
+        s = f'{char_name}: {greeting}'
+        t = f'**{char_name}**: {greeting}'
+        history_model = ['', s]
+        history_gradio = [('', t)]
+    return history_model, history_gradio, history_gradio, char_name, file_data["char_persona"], greeting, file_data["world_scenario"], file_data["example_dialogue"]
+
+def _build_character_settings_ui():
+
+    def char_file_create(char_name, char_persona, char_greeting, world_scenario, example_dialogue):
+        with open(char_name + ".json", "w") as f:
+            f.write(json.dumps({"char_name": char_name, "char_persona": char_persona, "char_greeting": char_greeting, "world_scenario": world_scenario, "example_dialogue": example_dialogue}))
+        return char_name + ".json"
+
+    with gr.Column():
+        with gr.Row():
+            char_name = gr.Textbox(
+                label="Character Name",
+                placeholder="The character's name",
+            )
+            user_name = gr.Textbox(
+                label="Your Name",
+                placeholder="How the character should call you",
+            )
+
+        char_persona = gr.Textbox(
+            label="Character Persona",
+            placeholder=
+            "Describe the character's persona here. Think of this as CharacterAI's description + definitions in one box.",
+            lines=4,
+        )
+        char_greeting = gr.Textbox(
+            label="Character Greeting",
+            placeholder=
+            "Write the character's greeting here. They will say this verbatim as their first response.",
+            lines=3,
+        )
+
+        world_scenario = gr.Textbox(
+            label="Scenario",
+            placeholder=
+            "Optionally, describe the starting scenario in a few short sentences.",
+        )
+        example_dialogue = gr.Textbox(
+            label="Example Chat",
+            placeholder=
+            "Optionally, write in an example chat here. This is useful for showing how the character should behave, for example.",
+            lines=4,
+        )
+
+        with gr.Row():
+            with gr.Column():
+                charfile = gr.File(type="binary", file_types=[".json"])
+
+                save_char_btn = gr.Button(value="Generate Character File")
+                save_char_btn.click(char_file_create, inputs=[char_name, char_persona, char_greeting, world_scenario, example_dialogue], outputs=[charfile])
+            with gr.Column():
+                gr.Markdown("""
+                    ### To save a character
+                    Click "Generate Character File". The file will appear above the button and you can click to download it.
+
+                    ### To upload a character
+                    Drag a valid .json file onto the upload box, or click the box to browse.
+                """)
+
+    return charfile, (char_name, user_name, char_persona, char_greeting, world_scenario, example_dialogue)
+
+
+def _build_generation_settings_ui(state, fn, for_kobold):
+    generation_defaults = get_generation_defaults(for_kobold=for_kobold)
+
+    with gr.Row():
+        with gr.Column():
+            max_new_tokens = gr.Slider(
+                16,
+                512,
+                value=generation_defaults["max_new_tokens"],
+                step=4,
+                label="max_new_tokens",
+            )
+            max_new_tokens.change(
+                lambda state, value: fn(state, "max_new_tokens", value),
+                inputs=[state, max_new_tokens],
+                outputs=state,
+            )
+
+            temperature = gr.Slider(
+                0.1,
+                2,
+                value=generation_defaults["temperature"],
+                step=0.01,
+                label="temperature",
+            )
+            temperature.change(
+                lambda state, value: fn(state, "temperature", value),
+                inputs=[state, temperature],
+                outputs=state,
+            )
+
+            top_p = gr.Slider(
+                0.0,
+                1.0,
+                value=generation_defaults["top_p"],
+                step=0.01,
+                label="top_p",
+            )
+            top_p.change(
+                lambda state, value: fn(state, "top_p", value),
+                inputs=[state, top_p],
+                outputs=state,
+            )
+
+        with gr.Column():
+            typical_p = gr.Slider(
+                0.0,
+                1.0,
+                value=generation_defaults["typical_p"],
+                step=0.01,
+                label="typical_p",
+            )
+            typical_p.change(
+                lambda state, value: fn(state, "typical_p", value),
+                inputs=[state, typical_p],
+                outputs=state,
+            )
+
+            repetition_penalty = gr.Slider(
+                1.0,
+                3.0,
+                value=generation_defaults["repetition_penalty"],
+                step=0.01,
+                label="repetition_penalty",
+            )
+            repetition_penalty.change(
+                lambda state, value: fn(state, "repetition_penalty", value),
+                inputs=[state, repetition_penalty],
+                outputs=state,
+            )
+
+            top_k = gr.Slider(
+                0,
+                100,
+                value=generation_defaults["top_k"],
+                step=1,
+                label="top_k",
+            )
+            top_k.change(
+                lambda state, value: fn(state, "top_k", value),
+                inputs=[state, top_k],
+                outputs=state,
+            )
+
+            if not for_kobold:
+                penalty_alpha = gr.Slider(
+                    0,
+                    1,
+                    value=generation_defaults["penalty_alpha"],
+                    step=0.05,
+                    label="penalty_alpha",
+                )
+                penalty_alpha.change(
+                    lambda state, value: fn(state, "penalty_alpha", value),
+                    inputs=[state, penalty_alpha],
+                    outputs=state,
+                )
+
+    #
+    # Some of these explanations are taken from Kobold:
+    # https://github.com/KoboldAI/KoboldAI-Client/blob/main/gensettings.py
+    #
+    # They're passed directly into the `generate` call, so they should exist here:
+    # https://huggingface.co/docs/transformers/main_classes/text_generation#transformers.GenerationConfig
+    #
+    with gr.Accordion(label="Helpful information", open=False):
+        gr.Markdown("""
+        Here's a basic rundown of each setting:
+
+        - `max_new_tokens`: Number of tokens the AI should generate. Higher numbers will take longer to generate.
+        - `temperature`: Randomness of sampling. High values can increase creativity but may make text less sensible. Lower values will make text more predictable but can become repetitious.
+        - `top_p`: Used to discard unlikely text in the sampling process. Lower values will make text more predictable but can become repetitious. (Put this value on 1 to disable its effect)
+        - `top_k`: Alternative sampling method, can be combined with top_p. The number of highest probability vocabulary tokens to keep for top-k-filtering. (Put this value on 0 to disable its effect)
+        - `typical_p`: Alternative sampling method described in the paper "Typical_p Decoding for Natural Language Generation" (10.48550/ARXIV.2202.00666). The paper suggests 0.2 as a good value for this setting. Set this setting to 1 to disable its effect.
+        - `repetition_penalty`: Used to penalize words that were already generated or belong to the context (Going over 1.2 breaks 6B models. Set to 1.0 to disable).
+        - `penalty_alpha`: The alpha coefficient when using contrastive search.
+
+        Some settings might not show up depending on which inference backend is being used.
+        """)

src/koboldai_client.py

@@ -0,0 +1,117 @@
+import datetime
+import logging
+import time
+
+import requests
+
+logger = logging.getLogger(__name__)
+
+
+class KoboldApiServerException(Exception):
+    pass
+
+
+def wait_for_kai_server(koboldai_url: str, max_wait_time_seconds: int) -> None:
+    '''Blocks until the KAI server is up.'''
+    start_time = datetime.datetime.now()
+
+    while True:
+        try:
+            requests.head(koboldai_url, timeout=(5, 5))
+            break
+        except requests.exceptions.ConnectionError as ex:
+            if "Connection refused" not in str(ex):
+                raise ex
+
+            abort_at = start_time + datetime.timedelta(
+                seconds=max_wait_time_seconds)
+
+            if datetime.datetime.now() > abort_at:
+                raise TimeoutError(
+                    f"Waited for {max_wait_time_seconds} seconds but KoboldAI"
+                    " server is still not up, aborting.")
+
+            time.sleep(1)
+
+
+def run_raw_inference_on_kai(
+    koboldai_url: str,
+    prompt: str,
+    max_new_tokens: int,
+    do_sample: bool,
+    typical_p: float,
+    repetition_penalty: float,
+    **kwargs,
+) -> str:
+    endpoint = f"{koboldai_url}/api/v1/generate"
+    payload = {
+        "prompt": prompt,
+
+        # Incredibly low max len for reasons explained in the "while True" loop
+        # below.
+        "max_length": 32,
+
+        # Take care of parameters which are named differently between Kobold and
+        # HuggingFace.
+        "sampler_full_determinism": not do_sample,
+        "typical": typical_p,
+        "rep_pen": repetition_penalty,
+
+        # Disable any pre or post-processing on the KoboldAI side, we'd rather
+        # take care of things on our own.
+        "frmttriminc": False,
+        "frmtrmspch": False,
+        "frmtrmblln": False,
+        "frmtadsnsp": False,
+
+        # Append any other generation parameters that we didn't handle manually.
+        **kwargs,
+    }
+    generated_text = ""
+
+    # Currently, Kobold doesn't support custom stopping criteria, and their chat
+    # mode can't handle multi-line responses. To work around both of those, we
+    # use the regular adventure mode generation but keep asking for more tokens
+    # until the model starts trying to talk as the user, then we stop.
+    attempts = 0
+    max_extra_attempts = 4
+    while attempts < (payload["max_length"] /
+                      max_new_tokens) + max_extra_attempts:
+        attempts += 1
+        response = requests.post(endpoint, json=payload)
+        if not response.ok:
+            error_message = response.text
+            raise KoboldApiServerException(
+                "The KoboldAI API server returned an error"
+                f" (HTTP status code {response.status_code}): {error_message}")
+
+        inference_result = response.json()["results"][0]["text"]
+        generated_text += inference_result
+
+        # Model started to talk as us. Stop generating and return results, the
+        # rest of the code will take care of trimming it properly.
+        if "\nYou:" in generated_text:
+            logger.debug("Hit `\nYou:`: `%s`", generated_text)
+            return generated_text
+
+        # For SFT: hit an EOS token. Trim and return.
+        if generated_text.endswith("<|endoftext|>"):
+            logger.debug("Got EOS token: `%s`", generated_text)
+
+            # We add a fake generated "\nYou:" here so the trimming code doesn't
+            # need to handle SFT and UFT models differently.
+            return generated_text.replace("<|endoftext|>", "\nYou:")
+
+        # Hit the configured generation limit.
+        if len(generated_text.split()) >= max_new_tokens:
+            logger.debug("Hit max length: `%s`", generated_text)
+            return generated_text
+
+        # Model still hasn't finished what it had to say. Append its output to
+        # the prompt and feed it back in.
+        logger.debug("Got another %s tokens, but still not done: `%s`",
+                     payload["max_length"], generated_text)
+        payload["prompt"] += inference_result
+
+    logger.debug("Exhausted generation attempts: `%s`", generated_text)
+    return generated_text

src/model.py

@@ -0,0 +1,110 @@
+import logging
+import typing as t
+
+import torch
+import transformers
+
+logger = logging.getLogger(__name__)
+
+
+def build_model_and_tokenizer_for(
+    model_name: str
+) -> t.Tuple[transformers.AutoModelForCausalLM, transformers.AutoTokenizer]:
+    '''Sets up the model and accompanying objects.'''
+    logger.info(f"Loading tokenizer for {model_name}")
+    tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
+
+    # NOTE(11b): non-OPT models support passing this in at inference time, might
+    # be worth refactoring for a debug version so we're able to experiment on
+    # the fly
+    bad_words_ids = [
+        tokenizer(bad_word, add_special_tokens=False).input_ids
+        for bad_word in _build_bad_words_list_for(model_name)
+    ]
+
+    logger.info(f"Loading the {model_name} model")
+    model = transformers.AutoModelForCausalLM.from_pretrained(
+        model_name, bad_words_ids=bad_words_ids)
+    model.eval().half().to("cuda")
+
+    logger.info("Model and tokenizer are ready")
+    return model, tokenizer
+
+
+def run_raw_inference(model: transformers.AutoModelForCausalLM,
+                      tokenizer: transformers.AutoTokenizer, prompt: str,
+                      user_message: str, **kwargs: t.Any) -> str:
+    '''
+    Runs inference on the model, and attempts to returns only the newly
+    generated text.
+
+    :param model: Model to perform inference with.
+    :param tokenizer: Tokenizer to tokenize input with.
+    :param prompt: Input to feed to the model.
+    :param user_message: The user's raw message, exactly as appended to the end
+        of `prompt`. Used for trimming the original input from the model output.
+    :return: Decoded model generation.
+    '''
+    tokenized_items = tokenizer(prompt, return_tensors="pt").to("cuda")
+
+    # Atrocious code to stop generation when the model outputs "\nYou: " in
+    # freshly generated text. Feel free to send in a PR if you know of a
+    # cleaner way to do this.
+    stopping_criteria_list = transformers.StoppingCriteriaList([
+        _SentinelTokenStoppingCriteria(
+            sentinel_token_ids=tokenizer(
+                "\nYou:",
+                add_special_tokens=False,
+                return_tensors="pt",
+            ).input_ids.to("cuda"),
+            starting_idx=tokenized_items.input_ids.shape[-1])
+    ])
+
+    logits = model.generate(stopping_criteria=stopping_criteria_list,
+                            **tokenized_items,
+                            **kwargs)
+    output = tokenizer.decode(logits[0], skip_special_tokens=True)
+
+    logger.debug("Before trimming, model output was: `%s`", output)
+
+    # Trim out the input prompt from the generated output.
+    if (idx := prompt.rfind(user_message)) != -1:
+        trimmed_output = output[idx + len(user_message) - 1:].strip()
+        logger.debug("After trimming, it became: `%s`", trimmed_output)
+
+        return trimmed_output
+    else:
+        raise Exception(
+            "Couldn't find user message in the model's output. What?")
+
+
+def _build_bad_words_list_for(_model_name: str) -> t.List[str]:
+    '''Builds a list of bad words for the given model.'''
+
+    # NOTE(11b): This was implemented as a function because each model size
+    # seems to have it quirks at the moment, but this is a rushed implementation
+    # so I'm not handling that, hence the dumb return here.
+    return ["Persona:", "Scenario:", "<START>"]
+
+
+class _SentinelTokenStoppingCriteria(transformers.StoppingCriteria):
+
+    def __init__(self, sentinel_token_ids: torch.LongTensor,
+                 starting_idx: int):
+        transformers.StoppingCriteria.__init__(self)
+        self.sentinel_token_ids = sentinel_token_ids
+        self.starting_idx = starting_idx
+
+    def __call__(self, input_ids: torch.LongTensor,
+                 _scores: torch.FloatTensor) -> bool:
+        for sample in input_ids:
+            trimmed_sample = sample[self.starting_idx:]
+            # Can't unfold, output is still too tiny. Skip.
+            if trimmed_sample.shape[-1] < self.sentinel_token_ids.shape[-1]:
+                continue
+
+            for window in trimmed_sample.unfold(
+                    0, self.sentinel_token_ids.shape[-1], 1):
+                if torch.all(torch.eq(self.sentinel_token_ids, window)):
+                    return True
+        return False

src/parsing.py

@@ -0,0 +1,40 @@
+import re
+import typing as t
+
+def parse_messages_from_str(string: str, names: t.List[str]) -> t.List[str]:
+    '''
+    Given a big string containing raw chat history, this function attempts to
+    parse it out into a list where each item is an individual message.
+    '''
+    sanitized_names = [
+        re.escape(name) for name in names
+    ]
+
+    speaker_regex = re.compile(rf"^({'|'.join(sanitized_names)}): ?",
+                               re.MULTILINE)
+
+    message_start_indexes = []
+    for match in speaker_regex.finditer(string):
+        message_start_indexes.append(match.start())
+
+    if len(message_start_indexes) < 2:
+        # Single message in the string.
+        return [string.strip()]
+
+    prev_start_idx = message_start_indexes[0]
+    messages = []
+
+    for start_idx in message_start_indexes[1:]:
+        message = string[prev_start_idx:start_idx].strip()
+        messages.append(message)
+        prev_start_idx = start_idx
+
+    # add the last message
+    messages.append(string[prev_start_idx:].strip())
+
+    return messages
+
+
+def serialize_chat_history(history: t.List[str]) -> str:
+    '''Given a structured chat history object, collapses it down to a string.'''
+    return "\n".join(history)

src/prompting.py

@@ -0,0 +1,54 @@
+import logging
+import typing as t
+
+from parsing import parse_messages_from_str
+
+logger = logging.getLogger(__name__)
+
+
+def build_prompt_for(
+    history: t.List[str],
+    user_message: str,
+    char_name: str,
+    char_persona: t.Optional[str] = None,
+    example_dialogue: t.Optional[str] = None,
+    world_scenario: t.Optional[str] = None,
+) -> str:
+    '''Converts all the given stuff into a proper input prompt for the model.'''
+
+    # If example dialogue is given, parse the history out from it and append
+    # that at the beginning of the dialogue history.
+    example_history = parse_messages_from_str(
+        example_dialogue, ["You", char_name]) if example_dialogue else []
+    concatenated_history = [*example_history, *history]
+
+    # Construct the base turns with the info we already have.
+    prompt_turns = [
+        # TODO(11b): Shouldn't be here on the original 350M.
+        "<START>",
+
+        # TODO(11b): Arbitrary limit. See if it's possible to vary this
+        # based on available context size and VRAM instead.
+        *concatenated_history[-8:],
+        f"You: {user_message}",
+        f"{char_name}:",
+    ]
+
+    # If we have a scenario or the character has a persona definition, add those
+    # to the beginning of the prompt.
+    if world_scenario:
+        prompt_turns.insert(
+            0,
+            f"Scenario: {world_scenario}",
+        )
+
+    if char_persona:
+        prompt_turns.insert(
+            0,
+            f"{char_name}'s Persona: {char_persona}",
+        )
+
+    # Done!
+    logger.debug("Constructed prompt is: `%s`", prompt_turns)
+    prompt_str = "\n".join(prompt_turns)
+    return prompt_str

src/utils.py

@@ -0,0 +1,11 @@
+def clear_stdout():
+    '''
+    Attempts to clear stdout, whether running in a notebook (IPython) or locally
+    in a Unix envirnoment.
+    '''
+    try:
+        from IPython.display import clear_output
+        clear_output(wait=True)
+    except ImportError:
+        import os
+        os.system("clear")