preferences
Change generation settings in the preferences pane (âď¸ icon in the upper right corner).
đ§ indicates dev-only settings, which can be changed if enable dev features is true.
global settings
These settings apply to all trees.
program
enable dev features
If checked, advanced features and features under development will be exposed. This includes exposing dev-only settings (indicated with đ§ on this page).
autosave to browser storage
If checked, the program state will be autosaved to the browser every 15 seconds.
More information: saving to browser storage
generation
đ§ use custom endpoint
Check to use a custom endpoint for model calls instead of one of the default models.
See custom model configs for how to configure calls to custom model endpoints.
expand tree on impossible navigation
If checked, new nodes will be generated if you attempt to navigate down using hotkeys (âĽS | alt+S) on a node without children or navigate left/right (âĽA | alt+A / âĽD | alt+D) when there is respectively no previous or next sibling.
This allows you to perform the main generation actions without having to press so many keys.
{video of workflow}
generate button branches from caret
Branch and generate from caret position when generate button pressed. If unchecked, generates from end of textbox.
đ§ branch on submit
map
show bookmark labels
If checked, bookmark label text will be shown on the map.
show text preview tooltips
If checked, node text preview tooltips will be shown when hovering on nodes in the map.
đ§ render tree
story textbox
show inline buttons
Toggle whether delete, bookmark, and sibling buttons are shown in the textbox when editing. Also controls whether branch buttons are shown in the textbox.
textbox writeable
Toggle whether the story textbox is editable. Disable to enable selecting/copying text across nodes.
tree-specific settings
These settings are saved with and may vary across individual tree files.
generation settings
model
The name of the model which generates.
Currently available through the Latitude API:
gpt-j
j1-jumbo (AI21)
j1-large (AI21)
You can also enable use custom endpoint to use other models. See custom model configs for how to configure calls to custom model endpoints.
# completions
Allowed values: 1-10
Number of parallel completions / child nodes to generate
max tokens
Allowed values: 1-1000
Maximum length of completions in tokens. One token is an average of about 3 characters.
temperature
Allowed values: 0-2:
Randomness of AI output. Higher temperatures for more varied and creative outputs. Lower temperatures for more predictable outputs. Too low temperature makes looping more likely.
stop sequences
up to 4 stop sequences allowed
The model will stop generating if it encounters any of these sequences. The stop sequence is not included in generated output.
memory
Memory is prepended to the prompt sent to the model (prompt = memory + story history). Subtracts from prompt length .
đ§ prompt length
default=5000
đ§ prepend to AI response
đ§ append to AI response
đ§ prepend to user action
đ§ append to user action
map settings
view
values ("full" or "pruned")
With enable dev features on, open filtering and pruning settings by clicking on the {filter} symbol above the map.
đ§ hidden tags
(default=['hidden', 'archived']
đ§ show only tags
(default=none)
đ§ collapse children
(default=true)
đ§ collapse chains
(default=false)
đ§ collapse chapters
(default=true)
Last updated