Ran into issues with other modes like chatmode and adventure, moved it further down the pipeline and converting </s> back to \n before processing additional formatting.
Still has an issue with the html formatting not working, but at least the AI works now.
On second thought, it is probably better to not save this. Advanced users can add this themselves and that way newer versions of the model can override it if redownloaded.
Allows model creators to customize the welcome message using Markdown and Limited HTML
Existing United users need to run install_requirements..bat again, you can leave the existing dependencies intact.
Adds a Nobreakmodel var that allows Breakmodel to be turned off. This can be done trough commandline or a model config (In case Neo is used by the models config without it being a true Neo model that is compatible with breakmodel).
In addition I removed the args.colab check for breakmodel support and instead make args.colab activate nobreakmodel. And I have added a new check so that breakmodel is not even attempted if you do not specify the layers but do launch a model from the command line.
Changed the model VRAM requirements to what you'd need to comfortably run the model rather than barely (Like with the manual). Will probably revise this in a later commit.
More importantly, it now supports models that use </s> which will be required to support XGLM and Fairseq models.
My last attempt at fixing this caused GPT2 to break, since the other fix is an edge case we assume that the GPT2 method should be used, and if that fails we try the other one to catch rare errors with bad model config's.
Turns out model_config does not work on models that have no model_type defined. In case this happens we now fall back to the old .json loading method. This will not work in --colab mode if its not already a local model, but since almost all modern models define a model type and to my knowledge all models on huggingface do that should not be an issue. If it is we can always ask the model creator to either update it, distribute the model differently or load that model with --remote instead of --colab.
In TPU instances, `vars.sp.shape[0]` is not always the actual number of
tokens in the soft prompt. We have to use `vars.sp_length` to get an
accurate token count.
Breakmodel is useless on Colab, so for the sake of efficiency if --colab is present we will always assume a model is incompatible. The same applies to the conversion, colab's are discarded so converting the model to a .bin file only wastes time since the HDD isn't fast. Finally we automatically set all the useful variables for Colab, so that in the future this can be removed from ckds and other scripts.
Lastly ckds has been adapted not to copy the examples folder and to add the new --colab parameter.
Local players are much better off running the old --remote command.
No longer update the chatname outside of the config, this will not effect singleplayer tab at all, but it will allow people in multiplayer to chat with their own names.
To prevent confusion with users who have not used KoboldAI for a while, or who are following old tutorials I have added a disclaimer that informs people that most Colab links should not be used with this feature and instead opened in the browser.
The problem was that when a soft prompt is being used, the dynamic
scanning criteria searches a different set of tokens for world info
keys than the `_generate()` function, which results in generation loops
when a world info key appears in the former set of tokens but not the
latter.
The `dynamic_processor_wrap` makes it so that the repetition penalty is
read directly from `vars`, but this only works if the initial repetition
sent to `generator` is not equal to 1. So we are now forcing the initial
repetition penalty to be something other than 1.
This commit exposes antemplates to the model config, this lets authors specify what kind of authors notes template they would like to use for their model. Users can still change it if they desire.
Blank lines appear often in chatmode so it is best played with blank line removal turned on, this is now forced. Its not compatible with Adventure mode, so they now turn each other off.
Added more models in the menu, all the popular community models are now easily accessible. I also re-ordered the menu from large to small to have it make a bit more sense.
* Error messages are now shown when memory, author's note, etc. exceeds
budget by itself
* Formatting options no longer break if there are empty chunks in the
story (although there shouldn't be any in the first place)
* Number of generated tokens is now kept track of from Python
* Removed `vars.model_orig`
* `requirex()` in bridge.lua now maintains a separate module cache for each
userscript instead of using the same cache for all userscripts
* `vars.lua_deleted` and `vars.lua_edited` are now erased right before running
the input modifiers instead of right before each time the generation modifiers
are run
The Initial commit for Chat Mode, the nickname part of the UI is missing other than that it should be fully functional. To use Chat Mode effectively you first input a small dialogue (Can be around 6 lines 3 of your own inputs and 3 of the character) formatted as Name : it will then automate the actions needed to chat properly. During this mode single line mode is forced on, and Trim Incomplete Sentences is forced off.
Futureproofing for future tokenizers, for now this is not needed since everything uses GPT2. But when that changes we want to be prepared. Not all models have a proper tokenizer config, so if we can't find one we fall back to GPT2.
First batch, will be more, we will also need to update the other VRAM display's with the changes that have happened. Will happen depending on how the 8-bit stuff goes.
NeoCustom is now obsolete beyond the file selection and the CLI. So after the CLI we adapt the input to a generic model and then use the improved generic routine to handle it. This saves duplicate efforts of maintaining an almost identical routine now that models are handled by their type and not their name.
This fixes a few scenario's of my commit yesterday, models that have a / are now first loaded from the corrected directory if it exists before we fall back to its original name to make sure it loads the config from the correct location. Cache dir fixes and a improved routine for the path loaded models that mimics the NeoCustom option fixing models that have no model_type specified. Because GPT2 doesn't work well with this option and should exclusively be used with the GPT2Custom and GPT-J models should have a model_type we assume its a Neo model when not specified.
Automatically detect or assume the model type so we do not have to hardcode all the different models people might use. This almost makes the behavior of --model identical to the NeoCustom behavior as far as the CLI is concerned. But only if the model_type is defined in the models config file.
Rather than coding a vars.custmodpath or vars.model in all the other parts of the code I opted to just set vars.custmodpath instead to make the behavior more consistent now that it always loads from the same location.
So that if you change, e.g., `top_p`, from a Lua generation modifier or
from the settings menu during generation, the rest of the generation
will use the new setting value instead of retaining the settings it had
when generation began.
Attempting to use transformers 4.11.0's experimental `low_cpu_mem_usage`
feature with GPT-2 models usually results in the output repeating a
token over and over or otherwise containing an incoherent response.
* `print()` and `warn()` now work correctly with `nil` arguments
* Typo: `gpt-neo-1.3M` has been corrected to `gpt-neo-1.3B`
* Regeneration is no longer triggered when writing to `keysecondary` of
a non-selective key
* Handle `genamt` changes in generation modifier properly
* Writing to `kobold.settings.numseqs` from a generation modifier no
longer affects
* Formatting options in `kobold.settings` have been fixed
* Added aliases for setting names
* Fix behaviour of editing story chunks from a generation modifier
* Warnings are now yellow instead of red
* kobold.logits is now the raw logits prior to being filtered, like
the documentation says, rather than after being filtered
* Some erroneous comments and error messages have been corrected
* These parts of the API have now been implemented properly:
* `compute_context()` methods
* `kobold.authorsnote`
* `kobold.restart_generation()`
Allow people to enter a prompt without generating anything by the AI. Combined with the always add prompt this is a very useful feature that allows people to write world information first, and then do a specific action. This mimics the behavior previously seen in AI Dungeon forks where it prompts for world information and then asks an action and can be particularly useful for people who want the prompt to always be part of the generation.
Automatically converts Huggingface cache models to full models on (down)load.
WARNING: Does wipe old cache/ dir inside the KoboldAI folder, make a backup before you run these models if you are bandwith constraint.
This seems to be related to the model config files, because only certain
models have this problem, and replacing ALL configuration files of a
"bad" model with those of a "good" model of the same type would fix the
problem.
Shouldn't be required anymore.
Recent optimizations caused the CPU version to load in an incompatible format, now we convert it back to the correct format after loading it efficiently first.
As requested by VE_FORBRYDERNE (Possibly implemented it on to many places, needs testing but since the other one is already broken I am committing it first so I can more easily test)
If the beginning of the comment is at the beginning of a line AND the
end of a comment is at the end of a line, an additional newline will now
be ignored so that the AI doesn't see a blank line where the comment
was.
For example, consider the following message:
```
Hello
<|This is
a comment|>
World
```
The AI will now see this:
```
Hello
World
```
instead of this:
```
Hello
World
```
Multiple things have changed, for now models default to half mode even on the official transformers to make sure its as efficient on the GPU as finetune's. GPU selection is streamlined and cache files are now stored inside the KoboldAI folder (for the most part). A new command line parameter to force the models to run at their full size still needs to be added for the few users that would want a quality bump at the cost of ram.
Changes the line-endings to the Unix format and sets KoboldAI to launch with Python3 if executed directly.
(cherry picked from commit 5b0977ceb6807c0f80ce6717891ef5e23c8eeb77)
The only changes are a small addition to the breakmodel section where GPU0 is automatically chosen if the CLI options are used without specifying breakmodel. Lineendings have been changed to Linux formatting for compatibility reasons.
Its made for Python3, so we assume python3 is installed in its usual location. If it isn't you can always run it yourself with whatever command you used prior to this change.
This prevents the "thinking" animation from appearing on top of the
submit button under certain circumstances:
* When someone connects to the KoboldAI server while the model is
generating (occurs after generation finishes)
* Occasionally, the browser may suddenly disconnect and reconnect from
Flask-SocketIO during generation, which causes the same problem
Apparently transformers maintains an internal reference to input_ids
(to use for repetition penalty) so we have to clamp the internal
version, too, because otherwise transformers will throw an out-of-bounds
error upon attempting to access token IDs that are not in the
vocabulary.
Adds Single Line mode, optimized for things like chatbot testing and other cases where you want to have control over what happens after a paragraph.
This can also be used as a foundation for a chatbot optimized interface mode.
breakmodel_layers and layers is confusing, changed the new method to breakmodel_gpulayers. The old one should no longer be used by people, but since it works in reverse we leave it in so scripts don't break.
Feedback from users is that its better to not always submit the prompt, this is consistent with the randomly generated stories. You can always toggle it on if you need this for coherency. This change does not override existing user settings.
Finetune's fork has unofficial support which we supported, but this is not compatible with models designed for the official version. In this update we let models decide which transformers backend to use, and fall back to Neo if they don't choose any. We also add the 6B to the menu and for the time being switch to the github version of transformers to be ahead of the waiting time. (Hopefully we can switch back to the conda version before merging upstream).
A new model was released that uses a different formatting for its enters, this causes to many enters in the UI. In this change we fix the issue so that when this happens the UI still displays the content as you would expect. Removing the formatting burden from the Model developers.
Originally omitted when model settings were forced. Now that models can only define the defaults for KoboldAI its a good idea to give model authors control over what formatting they think works best for their models.
If you save a story as a different name than it was loaded with, and
then try to download it as JSON/plaintext, the downloaded file's name
will now match the new story name.
This prevents duplicate submissions when multiple people are connected
to the same server and one person submits changes to memory, author's
note or world info, by pressing Submit (for author's note or memory) or
Accept (for world info).
Improved the default settings, better distinction on client / server. The python parts have been renamed to server, the browser to the client to be conform what you'd expect from a client and a server. The model name will also be shown now instead of NeoCustom.
Models can no longer override client settings, instead settings are now saved on a model per model basis with the settings provided by the model being the default. Users can also specify the desired configuration name as a command line parameter to avoid conflicting file names (Such as all Colabs having Colab.settings by default).
Many models have that one setting that just work best, like repetition penalty 2 or 1.2 while being incompatible with existing settings. Same applies for Adventure mode on or off. With this change models are allowed to override user preferences but only for the categories we deem this relevant (We don't want them to mess with things like tokens, length, etc). For users that do not want this behavior this can be turned off by changing msoverride to false in the client.settings.
Model creators can specify these settings in their config.json with the allowed settings being identical to their client.settings counterparts.
Bit of a workaround for now, but the [ badwords search routine has been replaced with a hardcoded list used by the colabs. This is far more effective at filtering out artifacts when running models locally. We can get away with this because all known models use the same vocab.json, in the future we will probably want to load this from badwords.json if present so model creators can bundle this with the model.
Replaces the placeholder readme with a proper one, the menu is also updated and reorganized to encourage users to use custom models and to better reflect the real world VRAM requirements.
Since some user interface buttons are disabled while in --remote mode,
they should also be disabled in aiserver.py so a malicious user can't
manually send those commands to the server.
Not just saving in .json but also in plain text, should help story writers get their stories out more easily. Especially since they can technically add some markdown into their stories manually in the interface.
* Back and Retry buttons no longer pop a story chunk while in the
"Select sequence to keep" menu
* No longer freezes if you retry with no story chunks beyond the initial
prompt chunk
* When "Always Add Prompt" is on, allow Retry even if the prompt is the
only chunk in the story
* Added error messages for Back and Retry buttons
Some colab's use KoboldAI as a subprocess, rather than making that to complicated for Colab developers its better to just dump the Cloudflare link to a log, in addition to showing the message on screen. That way if KoboldAI itself gets filtered you can easily cat the link or use the existing link grabbing methods.
henk717
ebolam
Gnome Ann
Ben Fox