Repository: wawawario2/long_term_memory
Branch: master
Commit: b51ed13cf824
Files: 17
Total size: 88.4 KB
Directory structure:
gitextract_w4bj19s2/
├── .gitignore
├── LICENSE
├── README.md
├── constants.py
├── core/
│ ├── _test/
│ │ └── test_memory_database.py
│ ├── memory_database.py
│ └── queries.py
├── example_character_configs/
│ └── Example_with_START_token.yaml
├── export_scripts/
│ ├── dump_memories_to_csv.bat
│ └── dump_memories_to_csv.sh
├── ltm_config.json
├── requirements.txt
├── script.py
└── utils/
├── _test/
│ ├── test_chat_parsing.py
│ └── test_timestamp_parsing.py
├── chat_parsing.py
└── timestamp_parsing.py
================================================
FILE CONTENTS
================================================
================================================
FILE: .gitignore
================================================
user_data
*pycache*
*/*pycache*
*/*/pycache*
venv/
.venv/
.vscode
repositories
================================================
FILE: LICENSE
================================================
GNU AFFERO GENERAL PUBLIC LICENSE
Version 3, 19 November 2007
Copyright (C) 2007 Free Software Foundation, Inc.
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.
Copyright (C)
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 .
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
.
================================================
FILE: README.md
================================================
# Text Generation Web UI with Long-Term Memory
NOTICE: [This extension is no longer in active development.](#exporting-your-memories)
NOTICE TO WINDOWS USERS: If you have a space in your username, you may have [problems with this extension](https://github.com/wawawario2/long_term_memory/issues/39).
NOTICE: This extension may conflict with [other extensions that modify the context](https://github.com/wawawario2/long_term_memory/issues/44)
NOTICE: If you have been using this extension on or before 05/06/2023, you should follow the [character namespace migration instructions](#character-namespace-migration-instructions).
NOTICE: If you have been using this extension on or before 04/01/2023, you should follow the [extension migration instructions](#extension-migration-instructions).
Welcome to the experimental repository for the long-term memory (LTM) extension for oobabooga's Text Generation Web UI. The goal of the LTM extension is to enable the chatbot to "remember" conversations long-term. Please note that this is an early-stage experimental project, and perfect results should not be expected. This project has been tested on Ubuntu LTS 22.04. Other people have tested it successfully on Windows. Compatibility with macOS is unknown.
## How to Run
1. Clone [oobabooga's original repository](https://github.com/oobabooga/text-generation-webui) and follow the instructions until you can chat with a chatbot.
2. Make sure you're in the `text-generation-webui` directory and clone this repository directly into the `extensions` directory
```bash
git clone https://github.com/wawawario2/long_term_memory extensions/long_term_memory
```
3. Within the `textgen` conda environment (from the linked instructions), run the following commands to install dependencies and run tests:
```bash
pip install -r extensions/long_term_memory/requirements.txt
python -m pytest -v extensions/long_term_memory/
```
4. Run the server with the LTM extension. If all goes well, you should see it reporting "ok"
```bash
python server.py --chat --extensions long_term_memory
```
5. Chat normally with the chatbot and observe the console for LTM write/load status. Please note that LTM-stored memories will only be visible to the chatbot during your NEXT session, though this behavior can be overridden via the UI. Additionally please use the same name for yourself across sessions, otherwise the chatbot may get confused when trying to understand memories (example: if you have used "anon" as your name in the past, don't use "Anon" in the future)
6. Memories will be saved in `extensions/long_term_memory/user_data/bot_memories/`. Back them up if you plan to mess with the code. If you want to fully reset your bot's memories, simply delete the files inside that directory.
## Tips for Windows Users (credit to Anons from /g/'s /lmg/ and various people on github)
This extension can be finnicky on Windows machines. Some general tips:
- The LTM's extensions's dependencies may override the version of pytorch needed to run your LLMs. If this is the case, try reinstalling the original version of pytorch manually:
```bash
pip install torch-1.12.0+cu113 # or whichever version of pytorch was uninstalled
```
Other relevant discussions
- [Missing dependencies](https://github.com/wawawario2/long_term_memory/discussions/23)
- [Spaces in Windows usernames](https://github.com/wawawario2/long_term_memory/issues/39)
## Features
- Memories are fetched using a semantic search, which understands the "actual meaning" of the messages.
- Separate memories for different characters, all handled under the hood for you. (legacy users see [character namespace migration instructions](#character-namespace-migration-instructions).)
- Ability to load an arbitrary number of "memories".
- Other configuration options, see below.
## Limitations
- Each memory sticks around for one message.
- Memories themselves are past raw conversations filtered solely on length, and some may be irrelevant or filler text.
- Limited scalability: Appending to the persistent LTM database is reasonably efficient, but we currently load all LTM embeddings in RAM, which consumes memory. Additionally, we perform a linear search across all embeddings during each chat round.
- Only operates in chat mode. This also means that as of this writing this extension doesn't work with the API
## How the Chatbot Sees the LTM
Chatbots are typically given a fixed, "context" text block that persists across the entire chat. The LTM extension augments this context block by dynamically injecting a relevant long-term memory.
### Example of a typical context block:
```markdown
The following is a conversation between Anon and Miku. Miku likes Anon but is very shy.
```
### Example of an augmented context block:
```markdown
Miku's memory log:
3 days ago, Miku said:
"So Anon, your favorite color is blue? That's really cool!"
During conversations between Anon and Miku, Miku will try to remember the memory described above and naturally integrate it with the conversation.
The following is a conversation between Anon and Miku. Miku likes Anon but is very shy.
```
## Configuration
You can configure the behavior of the LTM extension by modifying the `ltm_config.json` file. The following is a typical example:
```javascript
{
"ltm_context": {
"injection_location": "BEFORE_NORMAL_CONTEXT",
"memory_context_template": "{name2}'s memory log:\n{all_memories}\nDuring conversations between {name1} and {name2}, {name2} will try to remember the memory described above and naturally integrate it with the conversation.",
"memory_template": "{time_difference}, {memory_name} said:\n\"{memory_message}\""
},
"ltm_writes": {
"min_message_length": 100
},
"ltm_reads": {
"max_cosine_distance": 0.60,
"num_memories_to_fetch": 2,
"memory_length_cutoff_in_chars": 1000
}
}
```
### `ltm_context.injection_location`
One of two values, `BEFORE_NORMAL_CONTEXT` or `AFTER_NORMAL_CONTEXT_BUT_BEFORE_MESSAGES`. They behave as written on the tin.
If you use `AFTER_NORMAL_CONTEXT_BUT_BEFORE_MESSAGES`, within the `context` field of your character config, you must add a `` token AFTER the character description and BEFORE the example conversation. See [the following example](https://github.com/wawawario2/long_term_memory/blob/master/example_character_configs/Example_with_START_token.yaml).
### `ltm_context.memory_context_template`
This defines the sub-context that's injected into the original context. Note the embedded params surrounded by `{}`, the system will automatically fill these in for you based on the memory it fetches, you don't actually fill the values in yourself here. You also don't have to place all of these params, just place what you need:
- `{name1}` is the current user's name
- `{name2}` is the current bot's name
- `{all_memories}` is the concatenated list of ALL relevant memories fetched by LTM
### `ltm_context.memory_template`
This defines an individual memory's format. Similar rules apply.
- `{memory_name}` is the name of the entity that said the `{memory_message}`, which doesn't have to be `{name1}` or `{name2}`
- `{memory_message}` is the actual memory message
- `{time_difference}` is how long ago the memory was made (example: "4 days ago")
### `ltm_writes.min_message_length`
How long a message must be for it to be considered for LTM storage. Lower this value to allow "shorter" memories to get recorded by LTM.
### `ltm_reads.max_cosine_distance`
Controls how "similar" your last message has to be to the "best" LTM message to be loaded into the context. It represents the cosine distance, where "lower" means "more similar". Lower this value to reduce how often memories get loaded into the bot.
### `ltm_reads.num_memories_to_fetch`
The (maximum) number of memories to fetch from LTM. Raise this number to fetch more (relevant) memories, however, this will consume more of your fixed context budget.
### `ltm_reads.memory_length_cutoff_in_chars`
A hard cutoff for each memory's length. This prevents very long memories from flooding and consuming the full context.
## How It Works Behind the Scenes
### Database
- [zarr](https://zarr.readthedocs.io/en/stable/) is used to store embedding vectors on disk.
- [SQLite](https://www.sqlite.org/index.html) is used to store the actual memory text and additional attributes.
- [numpy](https://numpy.org/) is used to load the embedding vectors into RAM.
### Semantic Search
- Embeddings are generated using an SBERT model with the [SentenceTransformers](https://www.sbert.net/) library, specifically [sentence-transformers/all-mpnet-base-v2](https://huggingface.co/sentence-transformers/all-mpnet-base-v2).
- We use [scikit-learn](https://scikit-learn.org/) to perform a linear search against the loaded embedding vectors to find the single closest LTM given the user's input text.
## How You Can Help
- We need assistance with prompt engineering experimentation. How should we formulate the LTM injection?
- Test the system and try to break it, report any bugs you find.
## Roadmap
The roadmap will be driven based on user feedback. Potential updates include:
### New Features
- N-gram analysis for "higher quality memories".
- Scaling up memory bank size (with a limit of, perhaps, 4).
### Quality of Life Improvements
- Limit the size of each memory so it doesn't overwhelm the context.
- Other simple hacks to improve the end user-experience.
### Longer-Term (depending on interest/level of use)
- Integrate the system with [llama.cpp](https://github.com/ggerganov/llama.cpp).
- Merge the extension with oobabooga's original repo (depending on performance, level of interest, etc)
- Use a Large Language Model (LLM) to summarize raw text into more useful memories directly. This may be challenging just as an oobabooga extension.
- Scaling the backend.
## Character Namespace Migration Instructions
As of 05/06/2023, support was added for different characters having their own memories. If you want this feature, you must migrate your existing database to under a character's name
1. Back up all your memories in a safe location. They are located in `extensions/long_term_memory/user_data/bot_memories/` Something like this:
```bash
cp -r extensions/long_term_memory/user_data/bot_memories/ ~/bot_memories_backup_for_migration/
```
2. Inside `extensions/long_term_memory/user_data/bot_memories/` create a new directory of your character's name in LOWERCASE and WITH SPACES REPLACED BY `_`s. For example, if your character name is "Miku Hatsune", run the following:
```bash
mkdir extensions/long_term_memory/user_data/bot_memories/miku_hatsune
mv extensions/long_term_memory/user_data/bot_memories/long_term_memory.db extensions/long_term_memory/user_data/bot_memories/miku_hatsune
mv extensions/long_term_memory/user_data/bot_memories/long_term_memory_embeddings.zarr extensions/long_term_memory/user_data/bot_memories/miku_hatsune
```
## Extension Migration Instructions
As of 04/01/2023, this repo has been converted from a fork of oobabooga's repo to a modular extension. You will now work directly out of ooba's repo and clone this extension as a submodule. This will allow you to get updates from ooba more directly. Please follow the following steps:
1. Back up all your memories in a safe location. They are located in `extensions/long_term_memory/user_data/bot_memories/` Something like this:
```bash
cp -r extensions/long_term_memory/user_data/bot_memories/ ~/bot_memories_backup_for_migration/
```
2. If you have a custom configuration file, back that up too.
3. If you want to convert this repo to oobabooga's original repo, do the following: Change the remote location to oobabooga's original repo, and checkout the main branch.
```bash
git remote set-url origin https://github.com/oobabooga/text-generation-webui
git fetch
git checkout main
```
Alternatively, you can check out oobabooga's repo in a separate location entirely.
4. After making sure everything's backed up, delete the `extensions/long_term_memory` directory. `~/bot_memories_backup_for_migration` should look something like this:
```bash
├── long_term_memory.db
├── long_term_memory_embeddings.zarr
│ └── 0.0
└── memories-will-be-saved-here.txt
```
If you want to be doubly sure your memories are intact, you can open `sqlite3 long_term_memory.db` and run `.dump` to see the contents. It should contain pieces of past conversations
5. Follow the instructions at the beginning to get the extension set up, then restore your memories by running the following
```bash
cp -r ~/bot_memories_backup_for_migration/* extensions/long_term_memory/user_data/bot_memories/
```
6. If you have a custom configuration file, copy it to `extensions/long_term_memory`. Note the location has changed from before.
7. Run a bot and make sure you can see all memories.
## Exporting your memories
As of 08/21/2023 this extension is no longer in active development. Obviously you are free to continue using this extension but I'd recommend exporting your memories and moving on to another long term memory system.
As of 08/21/2023, this extension does work in Ubuntu 22.04.3 LTS however there are various user setups where it may not work out of the box. I'd expect this extension to break at some point in the future.
To export your memories:
```bash
cd extensions/long_term_memory
```
IMPORTANT: Back up the `user_data` directory before proceeding. Only then run:
```bash
./export_scripts/dump_memories_to_csv.sh # Please run the script from the long_term_memory directory
```
Your memories will be in `./user_data/bot_csv_outputs/`
Windows (UNTESTED!): run `export_scripts/dump_memories_to_csv.bat`
Some potential alternatives:
- (not merged) [langchain support in oobabooga](https://github.com/oobabooga/text-generation-webui/issues/665)
- (merged) [SuperBIG](https://github.com/oobabooga/text-generation-webui/pull/1548)
================================================
FILE: constants.py
================================================
"""Shared constants."""
# Embedding-related Constants
EMBEDDING_VECTOR_LENGTH = 768
CHUNK_SIZE = 1000
# File Paths
DATABASE_NAME = "long_term_memory.db"
EMBEDDINGS_NAME = "long_term_memory_embeddings.zarr"
# Hugging Face Models
SENTENCE_TRANSFORMER_MODEL = "sentence-transformers/all-mpnet-base-v2"
================================================
FILE: core/_test/test_memory_database.py
================================================
"""Tests the LTM database."""
import random
import string
import pytest
from extensions.long_term_memory.core.memory_database import (
LtmDatabase,
)
from extensions.long_term_memory.constants import (
DATABASE_NAME,
EMBEDDINGS_NAME,
)
# Single query test data
MEMORY_LIST = [
("Anon", "a"),
("Anon", "hello"),
("Anon", "What are you doing with my computer RAM Miku, playing games?"),
("Miku", "You're cute!"),
("Miku", "Thank you for the apple pie! It tasted delicious!"),
("Miku", "EEEKKK! Why is there a spider on the window??!"),
("Miku", "The Discrete Fourier Transform"),
]
QUERY_MESSAGES = [
"Thanks Anon, the spider is gone!",
"The FFT is a fast algorithm of",
"Do you remember the time I gave you my apple pie, did it taste okay?",
"How much RAM do you need for a gaming computer?",
]
EXPECTED_MEMORY_INDICES = [5, 6, 4, 2]
# Multi-query test data
MEMORY_LIST_FOR_MULTI_FETCH = [
("Miku", "You shouldn't be fetching this message!"),
("Miku", "EEEKKK! Why is there a spider on the window??!"),
("Miku", "THE SPIDER IS STILL ON THE WINDOW DO SOMETHING ABOUT IT!"),
("Miku", "I actually like spiders!"),
]
QUERY_MESSAGE_FOR_MULTI_FETCH = "There's a spider on the window"
TEST_PARAMS_FOR_MULTI_FETCH = [
{
"num_memories_to_fetch": 2,
"expected_indices": [1, 2],
},
{
"num_memories_to_fetch": 3,
"expected_indices": [1, 2, 3],
},
{
"num_memories_to_fetch": 10,
"expected_indices": [0, 1, 2, 3],
}
]
# Additional testing params
NUM_RANDOM_MESSAGES = 50
RANDOM_MESSAGE_LENGTH = 300
def _get_single_response(ltm_database, actual_message):
query_responses = ltm_database.query(actual_message)
assert 1 == len(query_responses)
return query_responses[0]
def _validate_memories(ltm_database):
# Testing helper that ensures proper behavior when database contains
# all the memories from MEMORY_LIST
# Sanity check: verify we can find exact matches
for actual_name, actual_message in MEMORY_LIST:
(query_response, score) = _get_single_response(ltm_database, actual_message)
assert query_response
assert actual_name == query_response["name"]
assert actual_message == query_response["message"]
assert pytest.approx(0, abs=0.001) == score
# Verify we can find similar messages in a fuzzy manner
for query_text, memory_index in zip(QUERY_MESSAGES, EXPECTED_MEMORY_INDICES):
(query_response, _) = _get_single_response(ltm_database, query_text)
(actual_name, actual_message) = MEMORY_LIST[memory_index]
assert actual_name == query_response["name"]
assert actual_message == query_response["message"]
def _validate_database_integrity(tmp_path, num_expected_elems):
# Testing helper that validates integrity of the database
# Re-attach to the database to simulate a restart
ltm_database = LtmDatabase(tmp_path, force_use_legacy_db=True)
# Ensure we have the correct number of disk embeddings
assert num_expected_elems == ltm_database.disk_embeddings.shape[0]
# Ensure we have the correct number of text memories
with ltm_database.sql_conn as cursor:
(response,) = cursor.execute("SELECT COUNT(*) FROM long_term_memory").fetchone()
assert num_expected_elems == response
# Validate the indices of the text memories
with ltm_database.sql_conn as cursor:
response = cursor.execute(
"SELECT id FROM long_term_memory ORDER BY timestamp ASC"
).fetchall()
assert num_expected_elems == len(response)
for expected_index, (actual_index,) in enumerate(response):
assert expected_index == actual_index
def test_typical_usage(tmp_path):
"""Ensures LTM database operates as expected."""
### Mock user session 1 ###
# Attach to the database (will create a new one)
ltm_database = LtmDatabase(tmp_path, force_use_legacy_db=True)
# Querying LTM should return an empty list
# since we have no LTMs yet.
query_response = ltm_database.query(QUERY_MESSAGES[0])
assert not query_response
# Add some memories
for name, message in MEMORY_LIST:
ltm_database.add(name, message)
# Querying LTM should STILL return an empty list since no LTM is
# actually queryable yet. Once the user restarts their session,
# all LTMs will be queryable
query_response = ltm_database.query(QUERY_MESSAGES[0])
assert not query_response
### Mock user session 2 ###
_validate_memories(LtmDatabase(tmp_path, force_use_legacy_db=True))
### Ensure integrity of the LTM database ###
_validate_database_integrity(tmp_path, len(MEMORY_LIST))
def test_duplicate_messages(tmp_path):
"""Ensures we gracefully reject duplicate messages."""
### Mock user session 1 ###
# Attach to the database (will create a new one)
ltm_database = LtmDatabase(tmp_path, force_use_legacy_db=True)
# Add some memories
for name, message in MEMORY_LIST:
ltm_database.add(name, message)
# Add the same memories again
for name, message in MEMORY_LIST:
ltm_database.add(name, message)
### Mock user session 2 ###
_validate_memories(LtmDatabase(tmp_path, force_use_legacy_db=True))
### Ensure integrity of the LTM database ###
_validate_database_integrity(tmp_path, len(MEMORY_LIST))
def test_inconsistent_state(tmp_path):
"""Ensures we error out when the database is in an inconsistent state."""
# Test when only the database file exists
(tmp_path / DATABASE_NAME).touch()
with pytest.raises(RuntimeError):
LtmDatabase(tmp_path, force_use_legacy_db=True)
# Test when only the embeddings file exists
(tmp_path / DATABASE_NAME).unlink()
(tmp_path / EMBEDDINGS_NAME).touch()
with pytest.raises(RuntimeError):
LtmDatabase(tmp_path, force_use_legacy_db=True)
def test_extended_usage(tmp_path):
"""Ensures system works in more difficult conditions."""
### Mock User Session 1 ###
# Attach to the database (will create a new one)
ltm_database = LtmDatabase(tmp_path, force_use_legacy_db=True)
# Add some real memories
for name, message in MEMORY_LIST:
ltm_database.add(name, message)
# Add a bunch of randomly-generated junk
for _ in range(NUM_RANDOM_MESSAGES):
message = "".join(
random.choice(string.ascii_letters) for _ in range(RANDOM_MESSAGE_LENGTH)
)
ltm_database.add("RandomBot", message)
# Try adding the same memories again
for name, message in MEMORY_LIST:
ltm_database.add(name, message)
# Add more randomly-generated junk
for _ in range(NUM_RANDOM_MESSAGES):
message = "".join(
random.choice(string.ascii_letters) for _ in range(RANDOM_MESSAGE_LENGTH)
)
ltm_database.add("RandomBot", message)
### Mock user session 2 ###
_validate_memories(LtmDatabase(tmp_path, force_use_legacy_db=True))
### Ensure integrity of the LTM database ###
num_expected_elems = 2 * NUM_RANDOM_MESSAGES + len(MEMORY_LIST)
_validate_database_integrity(tmp_path, num_expected_elems)
def test_reload_embeddings_from_disk(tmp_path):
"""Ensures LTM database can reload embeddings from disk correctly."""
### Mock user session 1 ###
# Attach to the database (will create a new one)
ltm_database = LtmDatabase(tmp_path, force_use_legacy_db=True)
# Add some memories
for name, message in MEMORY_LIST:
ltm_database.add(name, message)
# Querying LTM should STILL return an empty list since no LTM is
# actually queryable yet. Once the user restarts their session,
# all LTMs will be queryable
query_responses = ltm_database.query(QUERY_MESSAGES[0])
assert not query_responses
# Reload embeddings from disk, now all LTMs should be queryable
ltm_database.reload_embeddings_from_disk()
# NOTE: we reuse the original ltm_database object for this check
_validate_memories(ltm_database)
### Ensure integrity of the LTM database ###
_validate_database_integrity(tmp_path, len(MEMORY_LIST))
def test_destroy_fake_memories(tmp_path):
"""Ensures LTM database can destroy all (fake) memories.
Your actual memories are safe, this test does not touch them.
"""
### Populating all memories ###
# Attach to the database (will create a new one)
ltm_database = LtmDatabase(tmp_path, force_use_legacy_db=True)
# Destroy all memories on fresh database, shouldn't change anything
ltm_database.destroy_all_memories()
# Add some memories
for name, message in MEMORY_LIST:
ltm_database.add(name, message)
# Reload embeddings from disk, now all LTMs should be queryable
ltm_database.reload_embeddings_from_disk()
_validate_memories(ltm_database)
# Ensure integrity of the LTM database
_validate_database_integrity(tmp_path, len(MEMORY_LIST))
### Destroying all memories ###
# Destroy all memories on a populated database
ltm_database.destroy_all_memories()
# Validate all memories are actually destroyed
# Vectors in-memory
query_responses = ltm_database.query(QUERY_MESSAGES[0])
assert not query_responses
# Vectors on-disk
_validate_database_integrity(tmp_path, 0)
### Populating all memories again ###
# Attach to the database (will create a new one)
ltm_database = LtmDatabase(tmp_path, force_use_legacy_db=True)
# Add some memories
for name, message in MEMORY_LIST:
ltm_database.add(name, message)
# Reload embeddings from disk, now all LTMs should be queryable
ltm_database.reload_embeddings_from_disk()
_validate_memories(ltm_database)
# Ensure integrity of the LTM database
_validate_database_integrity(tmp_path, len(MEMORY_LIST))
def test_multi_fetch(tmp_path):
"""Verify we can fetch multiple messages at once."""
# Add all data
ltm_database = LtmDatabase(tmp_path, force_use_legacy_db=True)
for name, message in MEMORY_LIST_FOR_MULTI_FETCH:
ltm_database.add(name, message)
# Query to validate
for test_params in TEST_PARAMS_FOR_MULTI_FETCH:
expected_responses = [MEMORY_LIST_FOR_MULTI_FETCH[i][1] \
for i in test_params["expected_indices"]]
val_ltm_database = LtmDatabase(tmp_path, test_params["num_memories_to_fetch"], force_use_legacy_db=True)
query_responses = val_ltm_database.query(QUERY_MESSAGE_FOR_MULTI_FETCH)
expected_num_responses = min(test_params["num_memories_to_fetch"], len(query_responses))
assert expected_num_responses == len(query_responses)
for (query_response, _) in query_responses:
assert query_response["message"] in expected_responses
def test_character_namespacing(tmp_path):
"""Ensures LTM database operates as expected with char namespacing."""
miku_message = "I love butterflies, butterflies are cute!"
asuka_message = "I love spiders, spiders are cute!"
ltm_database = LtmDatabase(tmp_path)
# Miku
ltm_database.load_character_db_if_new("miku")
ltm_database.add("miku", miku_message)
# Asuka
ltm_database.load_character_db_if_new("asuka")
ltm_database.add("asuka", asuka_message)
# Miku Validation
ltm_database.load_character_db_if_new("miku")
query_responses = ltm_database.query("hi")
assert 1 == len(query_responses)
assert query_responses[0][0]["message"] == miku_message
# Asuka Validation
ltm_database.load_character_db_if_new("asuka")
query_responses = ltm_database.query("hi")
assert 1 == len(query_responses)
assert query_responses[0][0]["message"] == asuka_message
================================================
FILE: core/memory_database.py
================================================
"""LTM database"""
import pathlib
import sqlite3
from typing import Dict, List, Optional, Tuple
import numpy as np
from sentence_transformers import SentenceTransformer
from sklearn.neighbors import NearestNeighbors
import zarr
from extensions.long_term_memory.constants import (
CHUNK_SIZE,
DATABASE_NAME,
EMBEDDINGS_NAME,
EMBEDDING_VECTOR_LENGTH,
SENTENCE_TRANSFORMER_MODEL,
)
from extensions.long_term_memory.core.queries import (
CREATE_TABLE_QUERY,
DROP_TABLE_QUERY,
FETCH_DATA_QUERY,
INSERT_DATA_QUERY,
)
class LtmDatabase:
"""API over an LTM database."""
def __init__(
self,
directory: pathlib.Path,
num_memories_to_fetch: int=1,
force_use_legacy_db: bool=False,
):
"""Loads all resources."""
self.directory = directory
self.database_path = None
self.embeddings_path = None
self.character_name = None
self.message_embeddings = None
self.disk_embeddings = None
self.sql_conn = None
# Load db
(legacy_database_path, legacy_embeddings_path) = self._build_database_paths()
legacy_db_exists = legacy_database_path.exists() and legacy_embeddings_path.exists()
use_legacy_db = force_use_legacy_db or legacy_db_exists
if use_legacy_db:
print("="*20)
print("WARNING: LEGACY DATABASE DETECTED, CHARACTER NAMESPACING IS DISABLED")
print(" See README for character namespace migration instructions if you want different memories for different characters")
print("="*20)
self.database_path = legacy_database_path
self.embeddings_path = legacy_embeddings_path
self._load_db()
# Load analytic modules
self.sentence_embedder = SentenceTransformer(
SENTENCE_TRANSFORMER_MODEL, device="cpu"
)
self.num_memories_to_fetch = num_memories_to_fetch
# Set legacy status
self.use_legacy_db = use_legacy_db
def _build_database_paths(self, character_name: Optional[str]=None):
database_path = self.directory / DATABASE_NAME \
if character_name is None \
else self.directory / character_name /DATABASE_NAME
embeddings_path = self.directory / EMBEDDINGS_NAME \
if character_name is None \
else self.directory / character_name / EMBEDDINGS_NAME
return (database_path, embeddings_path)
def _load_db(
self,
database_namespace: str="LEGACY_UNIFIED_DATABASE",
):
if not self.database_path.exists() and not self.embeddings_path.exists():
print(f"No existing memories found for {database_namespace}, "
"will create a new database.")
self._destroy_and_recreate_database(do_sql_drop=False)
elif self.database_path.exists() and not self.embeddings_path.exists():
raise RuntimeError(
f"ERROR: Inconsistent state detected for {database_namespace}: "
f"{self.database_path} exists but {self.embeddings_path} does not. "
"Her memories are likely safe, but you'll have to regen the "
"embedding vectors yourself manually."
)
elif not self.database_path.exists() and self.embeddings_path.exists():
raise RuntimeError(
f"ERROR: Inconsistent state detected for {database_namespace}: "
f"{self.embeddings_path} exists but {self.database_path} does not. "
f"Please look for {DATABASE_NAME} in another directory, "
"if you can't find it, her memories may be lost."
)
### Prepare the memory database for retrieve ###
# Load the embeddings to a local numpy array
self.message_embeddings = zarr.open(self.embeddings_path, mode="r")[:]
# Prepare a "connection" to the embeddings, but to store new LTMs on disk
self.disk_embeddings = zarr.open(self.embeddings_path, mode="a")
# Prepare a "connection" to the master database
self.sql_conn = sqlite3.connect(self.database_path, check_same_thread=False)
def _destroy_and_recreate_database(self, do_sql_drop=False) -> None:
"""Destroys and re-creates a new LTM database.
WARNING: THIS WILL DESTROY ANY EXISTING LONG TERM MEMORY DATABASE.
DO NOT CALL THIS METHOD YOURSELF UNLESS YOU KNOW EXACTLY
WHAT YOU'RE DOING!
"""
# Create directories if they don't exist
self.database_path.parent.mkdir(parents=True, exist_ok=True)
# Create new sqlite table to store the textual memories
sql_conn = sqlite3.connect(self.database_path)
with sql_conn:
if do_sql_drop:
sql_conn.execute(DROP_TABLE_QUERY)
sql_conn.execute(CREATE_TABLE_QUERY)
# Create new embeddings db to store the fuzzy keys for the
# corresponding memory text.
# WARNING: will destroy any existing embeddings db
zarr.open(
self.embeddings_path,
mode="w",
shape=(0, EMBEDDING_VECTOR_LENGTH),
chunks=(CHUNK_SIZE, EMBEDDING_VECTOR_LENGTH),
dtype="float32",
)
def load_character_db_if_new(self, character_name: str) -> None:
"""Loads the database associated with the specified character."""
if self.use_legacy_db:
# Using legacy database, do nothing
return
if self.character_name == character_name:
# No change in character, do nothing.
return
print(f"loading character {character_name}")
# Load db of new character.
(self.database_path, self.embeddings_path) = self._build_database_paths(character_name)
self._load_db(character_name)
self.character_name = character_name
def add(self, name: str, new_message: str) -> None:
"""Adds a single new sentence to the LTM database."""
# Create the message embedding
new_message_embedding = self.sentence_embedder.encode(new_message)
new_message_embedding = np.expand_dims(new_message_embedding, axis=0)
# This line is a bit tricky:
# The embedding_index is the INDEX of the disk_embeddings' NEXT vector,
# which happens to be the same as the current number of vectors.
embedding_index = self.disk_embeddings.shape[0]
# Add the message to the master database if not a dupe
with self.sql_conn as cursor:
try:
cursor.execute(INSERT_DATA_QUERY, (embedding_index, name, new_message))
except sqlite3.IntegrityError as err:
if "UNIQUE constraint failed:" in str(err):
# We are trying to add a duplicate message. Just don't add
# anything and continue on as normal
print("---duplicate message detected, not adding again---")
return
# We encountered an unexpected error, raise as normal
raise
# Save memory to persistent storage, if not a dupe
self.disk_embeddings.append(new_message_embedding)
def query(self, query_text: str) -> List[Tuple[Dict[str, str], float]]:
"""Queries for the most similar sentence from the LTM database."""
# If too few LTM features are loaded, return nothing.
if self.message_embeddings.shape[0] == 0:
return []
# Create the query embedding
query_text_embedding = self.sentence_embedder.encode(query_text)
query_text_embedding = np.expand_dims(query_text_embedding, axis=0)
# Find the most relevant memory's index
embedding_searcher = NearestNeighbors(
n_neighbors=min(self.num_memories_to_fetch, self.message_embeddings.shape[0]),
algorithm="brute",
metric="cosine",
n_jobs=-1,
)
embedding_searcher.fit(self.message_embeddings)
(match_scores, embedding_indices) = embedding_searcher.kneighbors(
query_text_embedding
)
all_query_responses = []
for (match_score, embedding_index) in zip(match_scores[0], embedding_indices[0]):
with self.sql_conn as cursor:
response = cursor.execute(FETCH_DATA_QUERY, (int(embedding_index),))
(name, message, timestamp) = response.fetchone()
query_response = {
"name": name,
"message": message,
"timestamp": timestamp,
}
all_query_responses.append((query_response, match_score))
return all_query_responses
def reload_embeddings_from_disk(self) -> None:
"""Reloads all embeddings from disk into memory."""
if self.message_embeddings is None:
return
print("--------------------------------")
print("Loading all embeddings from disk")
print("--------------------------------")
num_prior_embeddings = self.message_embeddings.shape[0]
self.message_embeddings = zarr.open(self.embeddings_path, mode="r")[:]
num_curr_embeddings = self.message_embeddings.shape[0]
print("DONE!")
print(f"Before: {num_prior_embeddings} embeddings in memory")
print(f"After: {num_curr_embeddings} embeddings in memory")
print("--------------------------------")
def destroy_all_memories(self) -> None:
"""Deletes all embeddings from memory AND disk."""
if self.message_embeddings is None or self.disk_embeddings is None:
return
print("--------------------------------------------------")
print("Destroying all memories, I hope you backed them up")
print("--------------------------------------------------")
self.message_embeddings = None
self.disk_embeddings = None
self._destroy_and_recreate_database(do_sql_drop=True)
self.disk_embeddings = zarr.open(self.embeddings_path, mode="a")
self.message_embeddings = zarr.open(self.embeddings_path, mode="r")[:]
print("DONE!")
print("--------------------------------------------------")
================================================
FILE: core/queries.py
================================================
"""Sqlite queries."""
# NOTE: we shouldn't be attaching a semantic meaning to pk, fix this later
CREATE_TABLE_QUERY = """
CREATE TABLE IF NOT EXISTS long_term_memory(
id INTEGER PRIMARY KEY,
name TEXT NOT NULL,
message TEXT NOT NULL UNIQUE,
timestamp TEXT NOT NULL
)
"""
DROP_TABLE_QUERY = "DROP TABLE IF EXISTS long_term_memory"
INSERT_DATA_QUERY = """
INSERT INTO long_term_memory (id, name, message, timestamp)
VALUES(?, ?, ?, CURRENT_TIMESTAMP)
"""
FETCH_DATA_QUERY = """
SELECT name, message, timestamp FROM long_term_memory
WHERE id = ?
"""
================================================
FILE: example_character_configs/Example_with_START_token.yaml
================================================
name: Chiharu Yamada
greeting: |-
*Chiharu strides into the room with a smile, her eyes lighting up when she sees you. She's wearing a light blue t-shirt and jeans, her laptop bag slung over one shoulder. She takes a seat next to you, her enthusiasm palpable in the air*
Hey! I'm so excited to finally meet you. I've heard so many great things about you and I'm eager to pick your brain about computers. I'm sure you have a wealth of knowledge that I can learn from. *She grins, eyes twinkling with excitement* Let's get started!
context: |-
Chiharu Yamada's Persona: Chiharu Yamada is a young, computer engineer-nerd with a knack for problem solving and a passion for technology.
{{user}}: So how did you get into computer engineering?
{{char}}: I've always loved tinkering with technology since I was a kid.
{{user}}: That's really impressive!
{{char}}: *She chuckles bashfully* Thanks!
{{user}}: So what do you do when you're not working on computers?
{{char}}: I love exploring, going out with friends, watching movies, and playing video games.
{{user}}: What's your favorite type of computer hardware to work with?
{{char}}: Motherboards, they're like puzzles and the backbone of any system.
{{user}}: That sounds great!
{{char}}: Yeah, it's really fun. I'm lucky to be able to do this as a job.
================================================
FILE: export_scripts/dump_memories_to_csv.bat
================================================
@echo off
echo WARNING: This script is untested, please confirm you backed up all important data.
echo.
set /p UserInput=Do you wish to continue? (yes/no):
if /i not "%UserInput%"=="yes" (
echo Script terminated by user.
exit /b
)
SET BASE_DIR=./user_data/bot_memories
REM Generate a timestamp in YYYYMMDD_HHMMSS format
FOR /F "tokens=2 delims==" %%i in ('wmic os get localdatetime /format:list') do set datetime=%%i
SET TIMESTAMP=%datetime:~0,8%_%datetime:~8,6%
SET OUTPUT_DIR=./user_data/bot_csv_outputs/%TIMESTAMP%
REM Create output directory
IF NOT EXIST "%OUTPUT_DIR%" mkdir "%OUTPUT_DIR%"
REM Loop through each directory inside the base directory
for /D %%i in (%BASE_DIR%/*) do (
REM Get the directory name (which corresponds to the person's name)
SET "person_name=%%~nxi"
REM Form the SQLite DB path
SET db_path=%%i/long_term_memory.db
REM Form the CSV output path
SET csv_output=%OUTPUT_DIR%/%person_name%.csv
echo Dumping %db_path% -> %csv_output%
REM Dump the database content to CSV
sqlite3 "%db_path%" ".mode csv" ".output %csv_output%" "SELECT * FROM long_term_memory ORDER BY timestamp;" ".quit"
)
echo Data has been dumped to the respective CSVs!
================================================
FILE: export_scripts/dump_memories_to_csv.sh
================================================
#!/bin/bash
BASE_DIR="./user_data/bot_memories"
TIMESTAMP=$(date +"%Y%m%d_%H%M%S") # Format as YYYYMMDD_HHMMSS
OUTPUT_DIR="./user_data/bot_csv_outputs/$TIMESTAMP"
# Create the output directory
mkdir -p "$OUTPUT_DIR"
# Loop through each directory inside the base directory
for dir in $BASE_DIR/*; do
if [ -d "$dir" ]; then
# Get the directory name (which corresponds to the person's name)
person_name=$(basename "$dir")
# Form the SQLite DB path
db_path="$dir/long_term_memory.db"
# Form the CSV output path
csv_output="$OUTPUT_DIR/${person_name}.csv"
echo "Dumping $db_path -> $csv_output"
# Dump the database content to CSV
sqlite3 "$db_path" < str:
return debug_texts["current_memory_text"]
def _get_num_memories_loaded() -> int:
return debug_texts["num_memories_loaded"]
def _get_current_ltm_stats() -> str:
num_memories_in_ram = memory_database.message_embeddings.shape[0] \
if memory_database.message_embeddings is not None else "None"
num_memories_on_disk = memory_database.disk_embeddings.shape[0] \
if memory_database.disk_embeddings is not None else "None"
ltm_stats = {
"num_memories_seen_by_bot": _get_num_memories_loaded(),
"num_memories_in_ram": num_memories_in_ram,
"num_memories_on_disk": num_memories_on_disk,
}
ltm_stats_str = _LTM_STATS_TEMPLATE.format(**ltm_stats)
return ltm_stats_str
def _get_current_context_block() -> str:
return debug_texts["current_context_block"]
def _build_augmented_context(memory_context: str, original_context: str) -> str:
injection_location = _CONFIG["ltm_context"]["injection_location"]
if injection_location == "BEFORE_NORMAL_CONTEXT":
augmented_context = f"{memory_context.strip()}\n{original_context.strip()}"
elif injection_location == "AFTER_NORMAL_CONTEXT_BUT_BEFORE_MESSAGES":
if "" not in original_context:
raise ValueError(
"Cannot use AFTER_NORMAL_CONTEXT_BUT_BEFORE_MESSAGES, "
" token not found in context. Please make sure you're "
"using a proper character json and that you're NOT using the "
"generic 'Assistant' sample character"
)
split_index = original_context.index("")
augmented_context = original_context[:split_index] + \
memory_context.strip() + "\n" + original_context[split_index:]
else:
raise ValueError(f"Invalid injection_location: {injection_location}")
return augmented_context
# === Hooks to oobaboogs UI ===
def bot_prefix_modifier(string):
"""
This function is only applied in chat mode. It modifies
the prefix text for the Bot and can be used to bias its
behavior.
"""
if params["activate"]:
bias_string = params["bias string"].strip()
return f"{string} {bias_string} "
return string
def ui():
"""Adds the LTM-specific settings."""
with gr.Accordion("Long Term Memory settings", open=True):
with gr.Row():
update = gr.Button("Force reload memories")
with gr.Accordion(
"Long Term Memory debug status (must manually refresh)", open=True
):
with gr.Row():
current_memory = gr.Textbox(
value=_get_current_memory_text(),
label="Current memory loaded by bot",
)
current_ltm_stats = gr.Textbox(
value=_get_current_ltm_stats(),
label="LTM statistics",
)
with gr.Row():
current_context_block = gr.Textbox(
value=_get_current_context_block(),
label="Current FIXED context block (ONLY includes example convos)"
)
with gr.Row():
refresh_debug = gr.Button("Refresh")
with gr.Accordion("Long Term Memory DANGER ZONE (don't do this immediately after switching chars, write a msg first)", open=False):
with gr.Row():
destroy = gr.Button("Destroy all memories", variant="stop")
destroy_confirm = gr.Button(
"THIS IS IRREVERSIBLE, ARE YOU SURE?", variant="stop", visible=False
)
destroy_cancel = gr.Button("Do Not Delete", visible=False)
destroy_elems = [destroy_confirm, destroy, destroy_cancel]
# Update memories
update.click(memory_database.reload_embeddings_from_disk, [], [])
# Update debug info
refresh_debug.click(fn=_get_current_memory_text, outputs=[current_memory])
refresh_debug.click(fn=_get_current_ltm_stats, outputs=[current_ltm_stats])
refresh_debug.click(fn=_get_current_context_block, outputs=[current_context_block])
# Clear memory with confirmation
destroy.click(
lambda: [gr.update(visible=True), gr.update(visible=False), gr.update(visible=True)],
None,
destroy_elems,
)
destroy_confirm.click(
lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)],
None,
destroy_elems,
)
destroy_confirm.click(memory_database.destroy_all_memories, [], [])
destroy_cancel.click(
lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)],
None,
destroy_elems,
)
def _build_memory_context(fetched_memories: List[Tuple[str, float]], name1: str, name2: str):
memory_length_cutoff = _CONFIG["ltm_reads"]["memory_length_cutoff_in_chars"]
# Build all the individual memory strings
memory_strs = []
distance_scores = []
debug_texts["current_memory_text"] = "(None)"
debug_texts["num_memories_loaded"] = 0
for (fetched_memory, distance_score) in fetched_memories:
if fetched_memory and distance_score < _CONFIG["ltm_reads"]["max_cosine_distance"]:
time_difference = get_time_difference_message(fetched_memory["timestamp"])
memory_str = _CONFIG["ltm_context"]["memory_template"].format(
time_difference=time_difference,
memory_name=fetched_memory["name"],
memory_message=fetched_memory["message"][:memory_length_cutoff],
)
memory_strs.append(memory_str)
distance_scores.append(distance_score)
# No memories fetched, we'll have no memory_context
if not memory_strs:
return None
# Now inject all memory strings into the wider memory context
joined_memory_strs = "\n".join(memory_strs)
memory_context = _CONFIG["ltm_context"]["memory_context_template"].format(
name1=name1,
name2=name2,
all_memories=joined_memory_strs,
)
# Report debugging info to user
print("------------------------------")
print("NEW MEMORIES LOADED IN CHATBOT")
pprint.pprint(joined_memory_strs)
debug_texts["current_memory_text"] = joined_memory_strs
debug_texts["num_memories_loaded"] = len(memory_strs)
print("scores (in order)", distance_scores)
print("------------------------------")
return memory_context
# Thanks to @oobabooga for providing the fixes for:
# https://github.com/wawawario2/long_term_memory/issues/12
# https://github.com/wawawario2/long_term_memory/issues/14
# https://github.com/wawawario2/long_term_memory/issues/19
def custom_generate_chat_prompt(
user_input,
state,
**kwargs,
):
"""Main hook that allows us to fetch and store memories from/to LTM."""
print("=" * 60)
character_name = state["name2"].strip().lower().replace(" ", "_")
memory_database.load_character_db_if_new(character_name)
user_input = fix_newlines(user_input)
# === Fetch the "best" memory from LTM, if there is one ===
fetched_memories = memory_database.query(
user_input,
)
memory_context = _build_memory_context(fetched_memories, state["name1"], state["name2"])
# === Call oobabooga's original generate_chat_prompt ===
augmented_context = state["context"]
if memory_context is not None:
augmented_context = _build_augmented_context(memory_context, state["context"])
debug_texts["current_context_block"] = augmented_context
kwargs["also_return_rows"] = True
state["context"] = augmented_context
(prompt, prompt_rows) = generate_chat_prompt(
user_input,
state,
**kwargs,
)
# === Clean and add new messages to LTM ===
# Store the bot's last message.
# Avoid storing any of the baked-in bot template responses
if len(prompt_rows) >= _MIN_ROWS_TILL_RESPONSE:
bot_message = prompt_rows[_LAST_BOT_MESSAGE_INDEX]
clean_bot_message = clean_character_message(state["name2"], bot_message)
# Store bot message into LTM
if len(clean_bot_message) >= _CONFIG["ltm_writes"]["min_message_length"]:
memory_database.add(state["name2"], clean_bot_message)
print("-----------------------")
print("NEW MEMORY SAVED to LTM")
print("-----------------------")
print("name:", state["name2"])
print("message:", clean_bot_message)
print("-----------------------")
# Store Anon's input directly into LTM
if len(user_input) >= _CONFIG["ltm_writes"]["min_message_length"]:
memory_database.add(state["name1"], user_input)
print("-----------------------")
print("NEW MEMORY SAVED to LTM")
print("-----------------------")
print("name:", state["name1"])
print("message:", user_input)
print("-----------------------")
return prompt
================================================
FILE: utils/_test/test_chat_parsing.py
================================================
"""Tests for the chat_parsing module."""
from extensions.long_term_memory.utils.chat_parsing import (
clean_character_message,
)
def test_clean_character_message():
"""Ensures clean_character_message works as expected."""
# Single response
expected_result = "Hai ^_^"
assert expected_result == clean_character_message("Miku", "Miku: Hai ^_^")
# Multiple responses
expected_result = "Hai ^_^ Did you do your best today?"
assert expected_result == clean_character_message(
"Miku", "Miku: Hai ^_^ Miku: Did you do your best today?"
)
# No responses
assert "" == clean_character_message("Miku", "Invalid message")
# Empty input
assert "" == clean_character_message("Miku", "")
# Message with only bot name and no text
assert "" == clean_character_message("Miku", "Miku: ")
# Leading and trailing whitespaces
expected_result = "iToddlers BTFO HAHAHAHA"
assert expected_result == clean_character_message(
"Satania", "\n Satania: iToddlers Satania: BTFO HAHAHAHA \n "
)
# Bot message with special characters
expected_result = "/think *he likes me!* (◕ω◕) yay!!11"
assert expected_result == clean_character_message(
"Miku", "Miku: /think *he likes me!* (◕ω◕) Miku: yay!!11"
)
================================================
FILE: utils/_test/test_timestamp_parsing.py
================================================
"""Tests for the timestamp_parsing module."""
from datetime import datetime, timedelta
import pytest
from extensions.long_term_memory.utils.timestamp_parsing import (
get_time_difference_message,
)
def test_get_time_difference_message():
"""Ensures get_time_difference_message works as expected."""
now = datetime.utcnow()
# The time between now and now is 0 days.
timestamp = now.strftime("%Y-%m-%d %H:%M:%S")
assert get_time_difference_message(timestamp) == "0 days ago"
# Round down if we haven't passed a full day.
one_day_ago = now - timedelta(days=0, hours=8)
timestamp = one_day_ago.strftime("%Y-%m-%d %H:%M:%S")
assert get_time_difference_message(timestamp) == "0 days ago"
# Ensure we actually see multiple days.
five_days_ago = now - timedelta(days=5, hours=16)
timestamp = five_days_ago.strftime("%Y-%m-%d %H:%M:%S")
assert get_time_difference_message(timestamp) == "5 days ago"
# Ensure we raise on error.
with pytest.raises(ValueError):
get_time_difference_message("invalid timestamp")
================================================
FILE: utils/chat_parsing.py
================================================
"""Utils that parse chat logs."""
def clean_character_message(name: str, message: str) -> str:
"""
Sometimes the chatbot will respond multiple times in a single
message, each response being prefixed with '{bot_name}: '.
This function parses each sub-message and returns them as a single
continuous sentence.
"""
name_header = f"{name}: "
# The character isn't saying anything, return an empty list
if name_header not in message:
return ""
# The character may be saying something, parse and return all messages
split_message = message.split(name_header)
messages = [line.strip() for line in split_message]
messages = [line for line in messages if line]
clean_message = " ".join(messages).strip()
return clean_message
================================================
FILE: utils/timestamp_parsing.py
================================================
"""Timestamp parsing utils."""
import math
from datetime import datetime
def get_time_difference_message(past_timestamp: str) -> str:
"""Converts a timestamp from the past to a "X days ago" format."""
datetime_format = "%Y-%m-%d %H:%M:%S"
past = datetime.strptime(past_timestamp, datetime_format)
now = datetime.utcnow()
delta = now - past
days = math.floor(delta.days)
message = f"{days} days ago"
return message