kalomaze
2065f95edc
Sampler priority support
2024-02-10 14:57:41 -06:00
Cohee
d8d4732614
Add koboldcpp param aliases
2024-02-09 15:32:40 +02:00
kingbri
530455979f
Textgen: Add tabby to logprobs
...
Tabby now supports logprob returns. Add support to the backend gate.
Signed-off-by: kingbri <bdashore3@proton.me>
2024-02-08 20:59:54 -05:00
kingbri
f372b2ac16
Textgen: Fix dynatemp defaults
...
The default min and max temp values should be 1
Signed-off-by: kingbri <bdashore3@proton.me>
2024-02-08 20:57:40 -05:00
Cohee
f1cadfde2b
Fix llamacpp server
2024-02-04 17:41:46 +02:00
kingbri
3e34123b31
Textgen: Fix sampling param requests
...
1. TabbyAPI was not being included in temperature_last
2. DynaTemp parameters were still being sent even though the checkbox
was off. Cleanup the parameters if dynatemp is off for backends that
don't take in a dynatemp boolean.
Signed-off-by: kingbri <bdashore3@proton.me>
2024-02-04 01:44:07 -05:00
Cohee
6d0d92aaff
Add grammar and dynatemp for aphrodite
2024-02-02 20:06:46 +02:00
Alexander Abushady
da68d35ca5
Remove Noisy Sampling
...
Leaves DynaTemp Exponent and Quadratic Sampling.
2024-01-30 21:50:50 -05:00
Alexander Abushady
92aae0da04
Enter the Kalo Zone
2024-01-28 15:14:21 -05:00
AlpinDale
8af3939f09
fix: logprobs for Aphrodite
...
Since Aphrodite follows the same logprobs style as ooba, we can enable it easily.
2024-01-26 07:15:39 +00:00
khanon
60044c18a4
Implement Token Probabilities UI using logprobs
2024-01-25 18:34:46 -06:00
NWilson
f29f934c6b
Progress
2024-01-24 06:59:27 -06:00
Cohee
570d5a30bd
[skip ci] Lint fix
2024-01-20 20:40:40 +02:00
Cohee
5cb4dbc280
Trim TG types after split
2024-01-15 20:45:21 +02:00
Cohee
22e605168b
Properly reset seed with neutralize samplers
2024-01-15 20:44:07 +02:00
RossAscends
c6487fa9cb
make sampler nuke button work with zensliders
2024-01-14 18:33:40 +09:00
RossAscends
80b39bdac1
add neutralize samplers button for text completion slider panel
2024-01-14 18:07:43 +09:00
RossAscends
44bd57f763
filter ooba-only samplers out of non-ooba text completion slider panels
2024-01-14 17:26:22 +09:00
Cohee
1367642eb4
Fix maxTokens reference for n_predict alias
2024-01-12 12:20:03 +02:00
Cohee
8659bd57b7
Add koboldcpp rep pen aliases
2024-01-12 12:12:53 +02:00
Alexander Abushady
ec63cd8b6d
Dynatemp Range
...
Kobold Dynatemp range set when deactivated, now will work properly
2024-01-09 00:54:20 -05:00
Alexander Abushady
5ad980cf99
Fix for realzies
2024-01-09 00:02:53 -05:00
Alexander Abushady
04a5d8390d
Dynatemp UI v3.1
...
fixes for html positioning as well as api settings.
2024-01-08 23:58:06 -05:00
Alexander Abushady
b334acec23
DynaTemp UI v3
...
Backend bros won't play along now we gotta make them hold hands.
2024-01-08 22:20:27 -05:00
Cohee
2bea90a264
Merge pull request #1633 from AAbushady/new-dynatemp-ui
...
DynaTemp UI v2
2024-01-06 00:33:25 +02:00
Alexander Abushady
224dcc83a1
Update DynaTemp Sliders
...
More inline with changes made that will be implemented in KCPP main.
2024-01-05 14:36:39 -05:00
Cohee
5f93c30a96
#1627 Bypass status check and custom model for textgen type
2024-01-05 19:15:07 +02:00
Cohee
86d715cc16
Improve tokenizer detection
2024-01-05 16:17:06 +02:00
Alexander Abushady
d66be24518
DynaTemp UI v2
...
With DynaTemp moving into main KCPP now ST can have DynaTemp too!
2024-01-05 02:08:44 -05:00
Cohee
47cb017a45
#1589 Add 'cache_prompt' for l.cpp
2023-12-25 02:42:03 +02:00
Cohee
67dd52c21b
#1309 Ollama text completion backend
2023-12-19 16:38:11 +02:00
Cohee
edd737e8bd
#371 Add llama.cpp inference server support
2023-12-18 22:38:28 +02:00
Cohee
6e8104873e
#1569 Add logit bias for text completions
2023-12-18 18:57:10 +02:00
Cohee
b0d9f14534
Re-add Together as a text completion source
2023-12-17 23:38:03 +02:00
valadaptive
5b3c96df50
Rename /textgenerationwebui endpoint
...
I'd like to migrate over to using "textgen" to mean text-generation APIs
in general, so I've renamed the /textgenerationwebui/* endpoints to
/backends/text-completions/*.
2023-12-12 16:40:14 -05:00
Cohee
b0e7b73a32
Fix streaming processor error handler hooks
2023-12-08 02:01:08 +02:00
valadaptive
cdcd913805
Don't stream events if the API returned a 4xx code
2023-12-07 18:00:36 -05:00
valadaptive
5540c165cf
Refactor server-sent events parsing
...
Create one server-sent events stream class which implements the entire
spec (different line endings, chunking, etc) and use it in all the
streaming generators.
2023-12-07 18:00:36 -05:00
Cohee
0a092629ce
Fix undefined variable reference
2023-12-03 23:24:39 +02:00
valadaptive
24ccef3aba
Fix "settings" variable being shadowed
2023-12-03 15:44:36 -05:00
valadaptive
9c33ddbafc
Make textgen settings type checks more concise
2023-12-03 14:56:01 -05:00
valadaptive
047c897ead
Remove is[API] functions
...
Just use an equality comparison. It's a bit longer, but only because
"textgenerationwebui_settings" is a long identifier.
2023-12-03 14:56:01 -05:00
valadaptive
ba54e3dea0
Replaces is_[api] params with api_type param
...
These were 5 mutually-exclusive booleans, which can be replaced with one
param that takes on 5 values, one for each API type.
2023-12-03 14:56:01 -05:00
Cohee
939e938ba2
Disallow multiswipe for quiet gens
2023-12-03 20:56:25 +02:00
Cohee
1786b0d340
#1403 Add Aphrodite multi-swipe
2023-12-03 20:40:09 +02:00
Cohee
64a3564892
lint: Comma dangle
2023-12-02 22:06:57 +02:00
Cohee
c63cd87cc0
lint: Require semicolons
2023-12-02 21:11:06 +02:00
valadaptive
a37f874e38
Require single quotes
2023-12-02 13:04:51 -05:00
valadaptive
60ac1aa1c7
Enable no-empty lint
2023-12-02 10:32:07 -05:00
Cohee
61908935f5
Stop string for user-continue. Trim spaces after name2
2023-11-22 16:16:48 +02:00
Cohee
ddbdceba64
Add sampler order for koboldcpp under text completions
2023-11-21 01:47:57 +02:00
Cohee
1ebfddf07e
Use mistral and yi tokenizers for custom token bans
2023-11-21 01:04:27 +02:00
Cohee
96caddfd71
Add koboldcpp as Text Completion source
2023-11-19 17:14:53 +02:00
kingbri
f31b996cb5
Text Generation: Add TabbyAPI support
...
TabbyAPI is an exllamav2 only API server that aims to provide a simple
experience for loading and chatting with exl2 models.
SillyTavern currently doesn't have the ability to load and unload models,
so only add the OAI compatible completion endpoints.
The repository can be found here:
https://github.com/theroyallab/tabbyAPI
Signed-off-by: kingbri <bdashore3@proton.me>
2023-11-17 00:37:39 -05:00
Cohee
f4630f9808
Add lastMessage macro. Substitute params in CFG negative prompts
2023-11-16 16:58:49 +02:00
Cohee
61764a9a21
Change mancer base URL via debug menu
2023-11-13 11:13:39 +02:00
RossAscends
52c07e0895
setup Aphrodite-specific API flag handling
2023-11-13 16:36:01 +09:00
Cohee
480099ee97
Mancer will work in legacy API mode. Remove Soft Prompt mentions.
2023-11-08 18:16:47 +02:00
Cohee
738c1d82c4
Clarify legacy API notification
2023-11-08 17:54:08 +02:00
Cohee
cbc0b41773
Prevent doubled legacy toast
2023-11-08 10:28:02 +02:00
Cohee
b6d29d2980
Return force /api append for Legacy ooba with relaxed URL off
2023-11-08 10:25:49 +02:00
Cohee
e76c18c104
Legacy ooba API compatibility shim
2023-11-08 10:13:28 +02:00
RossAscends
00c029a65a
only api format for ooba, more visible checkboxes
2023-11-08 14:08:51 +09:00
Cohee
b2629d9718
Refactor status checks and indicators.
2023-11-08 02:52:03 +02:00
Cohee
bda15ef007
Adjust aphro streaming error parser
2023-11-08 00:25:06 +02:00
Cohee
2c7b954a8d
#1328 New API schema for ooba / mancer / aphrodite
2023-11-08 00:17:13 +02:00
RossAscends
a5b5dfdcb6
temperature last for ooba
2023-11-05 23:50:07 +09:00
RossAscends
937c5d7e5f
Min P for Ooba, fix preset loading for ZenSliders
2023-11-05 23:08:06 +09:00
RossAscends
447c43c9d7
placeholders for Aphro samplers
2023-11-04 11:16:12 +09:00
RossAscends
c79bf951b7
minor skill issue with slider overhaul
2023-10-26 20:41:12 +09:00
RossAscends
e9107870c0
Merge branch 'staging' of https://github.com/Cohee1207/SillyTavern into staging
2023-10-26 13:20:49 +09:00
RossAscends
d50124e937
appwide slider overhaul
2023-10-26 13:20:47 +09:00
kingbri
4f67e9f38b
Ooba: Add frequency and presence penalty sliders
...
These options essentially supersede repetition penalty, but since
ooba also gives the option, add these new sliders in conjunction
with the repetition penalty ones.
More testing is needed to find good default values.
Signed-off-by: kingbri <bdashore3@proton.me>
2023-10-25 15:22:47 -04:00
Cohee
c4fbc8373d
Add raw generate function
2023-10-11 17:56:52 +03:00
Cohee
d265179f46
Don't crash ST server on invalid streaming URL
2023-10-08 23:42:28 +03:00
valden80
788bbe969f
Feature, ability to add banned words sequences from anywhere through {{banned "..."}} macro. ( #1202 )
...
* Add files via upload
* Add files via upload
* Add files via upload
* Add files via upload
* Add files via upload
* Fix constant assignment, reformat code
---------
Co-authored-by: Cohee <18619528+Cohee1207@users.noreply.github.com>
2023-10-05 13:10:41 +03:00
Cohee
bb47712696
Split oobabooga/mancer sources. Add aphrodite support
2023-09-28 19:10:00 +03:00
Cohee
306cf51da4
#1180 Add custom token bans. Return grammar strings
2023-09-27 22:09:09 +03:00
Cohee
773d42b886
Suspend ooba grammar
2023-09-25 14:07:13 +03:00
Cohee
86e9974410
Add grammar for ooba llama.cpp
2023-09-24 17:57:00 +03:00
Cohee
322511caa9
Remove legacy Pygmalion formatting, part 2
2023-09-06 14:19:29 +03:00
Cohee
f48cc0db31
Unify cfgValues parsing between ooba/Novel
2023-08-23 18:26:56 +03:00
Cohee
fad6c164cb
Don't set negative prompt from CFG extension to ooba at guidance scale 1.0
2023-08-23 17:44:38 +03:00
Cohee
6923044a34
Allow CFG on impersonation
2023-08-22 19:35:56 +03:00
kingbri
25c7686462
CFG: Pass entire guidance object
...
If an API doesn't use the negative prompt from generate, allow
the negative prompt to be refetched in the specific API. This
requires the full guidance scale object instead of just the number.
Signed-off-by: kingbri <bdashore3@proton.me>
2023-08-22 10:17:20 -04:00
kingbri
92e6c6a998
CFG: Use ST prompt builder for negatives
...
Make the generate function build a negative prompt in addition to the
normal one. This allows for nonconflicting insertion with other extension
prompts and World Info.
Signed-off-by: kingbri <bdashore3@proton.me>
2023-08-20 00:44:39 -04:00
kingbri
7191f7a8ad
CFG: Add insertion depth and custom separator
...
Insertion depth allows for CFG to variably inject itself into the
negative prompt. This is similar to how Author's note works.
However, this method of insertion depth conflicts with AN and
world info where negatives can be meshed between two lines
of those specific insertions.
A custom separator must be wrapped in quotes, otherwise the default
separator is a newline for negative cascading.
Signed-off-by: kingbri <bdashore3@proton.me>
2023-08-19 01:29:59 -04:00
kingbri
cdbca6d9fd
CFG: Include the entire prompt with negative prompt
...
CFG with LLMs works differently than stable diffusion. The main principle
is prompt mixing and utilizing the differences between the two prompts
rather than a full "negative prompt" of what the user doesn't want.
SillyTavern its own way of formatting a prompt sent to an LLM backend.
Therefore, take that prompt and add negatives to it.
Signed-off-by: kingbri <bdashore3@proton.me>
2023-08-18 22:18:09 -04:00
Cohee
32eb877fdb
Add native presetable CFG scale for ooba
2023-08-14 13:06:20 +03:00
kingbri
5bb6c89868
CFG: Add per-chat CFG and fixes
...
Per-chat CFG applies a CFG setting per-chat only rather than character
or globally. This overrides all other CFG settings (this will be changed).
Also add fixes to remove character CFG entries properly and not to
apply CFG if the scale is 1 as that won't do anything to generation.
Signed-off-by: kingbri <bdashore3@proton.me>
2023-08-11 01:39:15 -04:00
kingbri
0c94740560
Extensions: Add global CFG support
...
CFG is bundled as an extension which is more flexible in terms of
storing settings and consolidating code.
Global CFG is currently supported and the old power user setting
is auto-migrated.
Signed-off-by: kingbri <bdashore3@proton.me>
2023-08-11 01:39:15 -04:00
50h100a
53e41bdda8
Relax URL requirements when Mancer is enabled.
2023-08-08 16:12:03 -04:00
50h100a
c1ab0212e5
update comments
2023-08-07 17:10:05 -04:00
50h100a
128945aaaa
Copy 'relaxed api url' functionality over to webui
2023-08-07 16:46:32 -04:00
50h100a
2fc6813e66
code review adjustments part 1
2023-08-03 06:07:54 -04:00
Cohee
130559d499
#746 Add preset manager for ooba/kobold
2023-07-23 23:52:31 +03:00
Cohee
1b9ca4c9d9
Unhide advanced parameters for Novel
2023-07-23 02:09:03 +03:00
ThisIsPIRI
e2f1f4a57a
Add repetition penalty range for ooba
2023-07-22 07:38:35 +09:00
Cohee
edd41989fd
Initial commit
2023-07-20 20:32:15 +03:00
