80256012ad
* fix issue where saving a new scene would save into a "new scenario" directory instead instead of a relevantly named directory * implement function to fork new scene file from specific message * dynamic choice generation * dynamic choice generation progress * prompt tweaks * disable choice generation by default prompt tweaks * prompt tweaks for assisted RAG tasks * allow analyze_text_and_extract_context to include character context * more prompt tweaks for RAG assist during conversation generation * open director settings from dynamic action dialog * adjust wording * remove player choice message if the trigger message is removed (or regenerated) * fix issue with dialogue cleaqup where narration over multiple lines would end up being marked incorrectly * dynamic action generation custom instructions dynamic action generation narration for sensory actions * fix actions when acting as another character * 0.28.0 * conversation agent: split out generation settings, add actor instructions extension, add actor instruction offset slider * prompt tweaks * fix ai message regenerate if generated from choice * cruft * layered history implementation through summarizer summarization tweaks * show layered history in ux * layered history fixes and tweaks conversation actor instruction fixes * more summarization fixes * fix missing actor instructions * prompt tweaks * prompt tweaks * force lower case when checking sensory type * agent modal polish implement find-natural-scene-termination summarizer action some summarization tweaks * integrate find_natural_scene_termination with layered history * collect all denouements at once * relock * fix some issues with screenplay type formatting in conversation agent * cleanup * revert layered history summarization to use max_process_tokens instead of using ai to fine scene termination as that process falls apart in layer 1 and higher, at that point every item is a scene in itself. * implement ai assisted digging through layered history to answer queries * dig_layered_history tweaks and improvements * prompt tweaks * adjust budget * adjust budget for RAG context * layered_history disabled by default * prompt tweaks to reinforcement updates * prompt tweaks * dig layered history - response without function call to be treated as answer * clarify style keywords to avoid bleeding into the prompt as subject matter * fix issue with cover image updates * fix missing dialogue from context history * fix issue where new scenes wouldn't load * fix crash with layered summarization * more context history fixes * fix assured dialogue message in context history * prompt tweaks * tweaks to layered history generation * prompt tweaks * conversation agent can dig layered history for extra context * some fixes to dig layered history * scene fork adjust layered history * layered history status indication * allow configuration of message styles and colors * fix issue where layered history generate would get stuck on layer 0 * dig layered history default to false * prompt tweaks * context investigation messages * tweaks to context investigation * context investigation polish of UX and allow specifying trigger * prompt tweaks * allow hiding of ci and director messages * wire ci shrotcut buttons * prompt tweaks * prompt tweaks * carry on analysis when digging layered history * improve quality of generate choices by anchoring to last line in the scene * update hint message * prompt tweaks * change default value for max_process_tokens * docs * dig layered history only if there are layers * always enforce num choices limit * relock * typos * prompt tweaks * docs for forking a scene * prompt tweaks * world editor rubber banding fixes follow up * layered history cleanup fixes * gracefully handle malformed dig() call * handle malformed answer() call * only generate choices if last content isn't player message * include more context in autocomplete prompts * prompt tweaks * typo * fix issue where inactive characters could not be deleted * more character delete bugs * dig layered history fixes * discard empty content investigations * fix issue with autocomplete no longer working in world editor * prompt tweaks * support single quotes * prompt tweaks * fix issue with context investigation if final message was narrator text * Include the query in the context investigation message * context investigvations should note when historic events occured * instructions on how to use internal notes * time_diff return empty string no time supplied * prompt tweaks * fix date calculations for historic entries * change default values * prompt tweaks * fix history regenerate continuing through page reload * reorganize websocket tasks * allow cancelling of history regenerate * Capitalize first letter of summarization * include base layer in context investigations * prompt tweaks * fix issue where context investigations would expand too much of the history at once * attempt to determine character knowledge during context investigation * prompt tweaks * prompt tweaks * fix mising timestamps * more context during layer history digging * fix issue with act-as not being able to select past the first npc if a scene had more than one active npcs in it * docs * error handling for malformed answer call * timestamp calculation fixes and summarization improvements * lock message manipulation while the ux is busy * prompt tweaks * toggling 'log debug messages' will log all messages to console even if no filter is specified * layered history generation cancellable from ux * prevent loading scene while another scene is currently loading * improvements to choice generation prompt and error handling * prompt tweaks * prompt tweaks * prompt tweaks * fix issue with successive scene load not working * correctly display timestamps and generated layers during history regen * summarization improvements * clean up context investigation prompt * prompt tweaks * increase response token size for dig_layered_history * define missing presets * missing preset * prompt tweaks * fix simulation suite * attach punkt download to backend start, not frontend start * dig layered history fixes * prompt tweaks * fix summarize_and_pin * more fixes for time calculations * relock * prompt tweaks * remove dupe entry from layered history * bash version of update script * prompt tweaks * layered history defaults to enabled * default decreased to 0.3 chance * fix multi character natural flow selection with clients that don't support LLM coercion * fix simulation suite call to change a character * typo * remove deprecated test * use python3 * add missing 4o models * add proper configs for 4o models * prompt tweaks * update reinforcement prompt ignores context investigations * scene.snapshot formatting and dig_layered_history ignores reinforcments * use end date instead of start date * Reword 'Moments ago' to 'Recently' as it is more forgiving and applicable to longer time ranges * fix time calculation issues during summarization of new entries * no need for scoping * dont display as range if start and end of entry are identical * prompt tweaks
2.8 KiB
Settings
General
General summarization settings.
Summarize to long term memory archive
Automatically summarize scene dialogue when the number of tokens in the history exceeds a threshold. This helps keep the context history from growing too large.
Token threshold
The number of tokens in the history that will trigger the summarization process.
Summarization method
The method used to summarize the scene dialogue.
Balanced- medium length summaryShort & Concise- short summaryLengthy & Detailed- long summaryFactual list- numbered list of events that transpired
Use preceeding summaries to strengthen context
Help the AI summarize by including the last few summaries as additional context. Some models may incorporate this context into the new summary directly, so if you find yourself with a bunch of similar history entries, try setting this to 0.
Layered History
Settings for the layered history summarization.
Talemate 0.28.0 introduces a new feature called layered history summarization. This feature allows the AI to summarize the scene dialogue in layers, with each layer providing a different level of detail.
Not only does this allow to keep more context in the history, albeit with earlier layers containing less detail, but it also allows us to do history investgations to extract relevant information from the history during conversation and narration prompts.
Right now this is considered an experimental feature, and whether or not its feasible in the long term will depend on how well it works in practice.
Enable layered history
Allows you to enable or disable the layered history summarization.
!!! note "Enabling this on big scenes" If you enable this on a big established scene, the next time the summarization agent runs, it will take a while to process the entire history and generate the layers.
Token threshold
The number of tokens in the layer that will trigger the summarization process to the next layer.
Maximum number of layers
The maximum number of layers that can be created. Raising this limit past 3 is likely to have dimishing returns. We have observed that usually by layer 3 you are down to single sentences for individual events, making it difficult to summarize further in a meaningful way.
Maximum tokens to process
Smaller LLMs may struggle with accurately summarizing long texts. This setting will split the text into chunks and summarize each chunk separately, then stitch them together in the next layer. If you're using a strong LLM (70B+), you can try setting this to be the same as the threshold.
Setting this higher than the token threshold does nothing.