talemate/scenes/simulation-suite/game.py
veguAI 80256012ad
0.28.0 (#148)
* 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
2024-11-24 15:43:27 +02:00

594 lines
No EOL
27 KiB
Python

def game(TM):
MSG_PROCESSED_INSTRUCTIONS = "Simulation suite processed instructions"
MSG_HELP = "Instructions to the simulation computer are only processed if the computer is directly addressed at the beginning of the instruction. Please state your commands by addressing the computer by stating \"Computer,\" followed by an instruction. For example ... \"Computer, i want to experience being on a derelict spaceship.\""
PROMPT_NARRATE_ROUND = "Narrate the simulation and reveal some new details to the player in one paragraph. YOU MUST NOT ADDRESS THE COMPUTER OR THE SIMULATION."
PROMPT_STARTUP = "Narrate the computer asking the user to state the nature of their desired simulation in a synthetic and soft sounding voice."
CTX_PIN_UNAWARE = "Characters in the simulation ARE NOT AWARE OF THE COMPUTER OR THE SIMULATION."
AUTO_NARRATE_INTERVAL = 10
def parse_sim_call_arguments(call:str) -> str:
"""
Returns the value between the parentheses of a simulation call
Example:
call = 'change_environment("a house")'
parse_sim_call_arguments(call) -> "a house"
"""
try:
return call.split("(", 1)[1].split(")")[0]
except Exception:
return ""
class SimulationSuite:
def __init__(self):
"""
This is initialized at the beginning of each round of the simulation suite
"""
# do we update the world state at the end of the round
self.update_world_state = False
self.simulation_reset = False
# will keep track of any npcs added during the current round
self.added_npcs = []
TM.log.debug("SIMULATION SUITE INIT!", scene=TM.scene)
self.player_message = TM.scene.last_player_message
self.last_processed_call = TM.game_state.get_var("instr.lastprocessed_call", -1)
# determine whether the player / user input is an instruction
# to the simulation computer
#
# we do this by checking if the message starts with "Computer,"
self.player_message_is_instruction = (
self.player_message and
self.player_message.raw.lower().startswith("computer") and
not self.player_message.hidden and
not self.last_processed_call > self.player_message.id
)
def run(self):
"""
Main entry point for the simulation suite
"""
if not TM.game_state.has_var("instr.simulation_stopped"):
# simulation is still running
self.simulation()
self.finalize_round()
def simulation(self):
"""
Simulation suite logic
"""
if not TM.game_state.has_var("instr.simulation_started"):
self.startup()
else:
self.simulation_calls()
if self.update_world_state:
self.run_update_world_state(force=True)
def startup(self):
"""
Scene startup logic
"""
# we are at the beginning of the simulation
TM.signals.status("busy", "Simulation suite powering up.", as_scene_message=True)
TM.game_state.set_var("instr.simulation_started", "yes", commit=False)
# add narration for the introduction
TM.agents.narrator.action_to_narration(
action_name="progress_story",
narrative_direction=PROMPT_STARTUP,
emit_message=False
)
# add narration for the instructions on how to interact with the simulation
# this is a passthrough since we don't want the AI to paraphrase this
TM.agents.narrator.action_to_narration(
action_name="passthrough",
narration=MSG_HELP
)
# create a world state entry letting the AI know that characters
# interacting in the simulation are not aware of the computer or the simulation
TM.agents.world_state.save_world_entry(
entry_id="sim.quarantined",
text=CTX_PIN_UNAWARE,
meta={},
# this should always be pinned
pin=True
)
# set flag that we have started the simulation
TM.game_state.set_var("instr.simulation_started", "yes", commit=False)
# signal to the UX that the simulation suite is ready
TM.signals.status("success", "Simulation suite ready", as_scene_message=True)
# we want to update the world state at the end of the round
self.update_world_state = True
def simulation_calls(self):
"""
Calls the simulation suite main prompt to determine the appropriate
simulation calls
"""
# we only process instructions that are not hidden and are not the last processed call
if not self.player_message_is_instruction or self.player_message.id == self.last_processed_call:
return
# First instruction?
if not TM.game_state.has_var("instr.has_issued_instructions"):
# determine the context of the simulation
context_context = TM.agents.creator.determine_content_context_for_description(
description=self.player_message.raw,
)
TM.scene.set_content_context(context_context)
# Render the `computer` template and send it to the LLM for processing
# The LLM will return a list of calls that the simulation suite will process
# The calls are pseudo code that the simulation suite will interpret and execute
calls = TM.prompt.request(
"computer",
dedupe_enabled=False,
player_instruction=self.player_message.raw,
scene=TM.scene,
)
self.calls = calls = calls.split("\n")
calls = self.prepare_calls(calls)
TM.log.debug("SIMULATION SUITE CALLS", callse=calls)
# calls that are processed
processed = []
for call in calls:
processed_call = self.process_call(call)
if processed_call:
processed.append(processed_call)
if processed:
TM.log.debug("SIMULATION SUITE CALLS", calls=processed)
TM.game_state.set_var("instr.has_issued_instructions", "yes", commit=False)
TM.signals.status("busy", "Simulation suite altering environment.", as_scene_message=True)
compiled = "\n".join(processed)
if not self.simulation_reset and compiled:
# send the compiled calls to the narrator to generate a narrative based
# on them
narration = TM.agents.narrator.action_to_narration(
action_name="progress_story",
narrative_direction=f"The computer calls the following functions:\n\n```\n{compiled}\n```\n\nand the simulation adjusts the environment according to the user's wishes.\n\nWrite the narrative that describes the changes to the player in the context of the simulation starting up. YOU MUST NOT REFERENCE THE COMPUTER OR THE SIMULATION.",
emit_message=True
)
# on the first narration we update the scene description and remove any mention of the computer
# or the simulation from the previous narration
is_initial_narration = TM.game_state.get_var("instr.intro_narration", False)
if not is_initial_narration:
TM.scene.set_description(narration.raw)
TM.scene.set_intro(narration.raw)
TM.log.debug("SIMULATION SUITE: initial narration", intro=narration.raw)
TM.scene.pop_history(typ="narrator", all=True, reverse=True)
TM.scene.pop_history(typ="director", all=True, reverse=True)
TM.game_state.set_var("instr.intro_narration", True, commit=False)
self.update_world_state = True
self.set_simulation_title(compiled)
def set_simulation_title(self, compiled_calls):
"""
Generates a fitting title for the simulation based on the user's instructions
"""
TM.log.debug("SIMULATION SUITE: set simulation title", name=TM.scene.title, compiled_calls=compiled_calls)
if not compiled_calls:
return
if TM.scene.title != "Simulation Suite":
# name already changed, no need to do it again
return
title = TM.agents.creator.contextual_generate_from_args(
"scene:simulation title",
"Create a fitting title for the simulated scenario that the user has requested. You response MUST be a short but exciting, descriptive title.",
length=75
)
title = title.strip('"').strip()
TM.scene.set_title(title)
def prepare_calls(self, calls):
"""
Loops through calls and if a `set_player_name` call and a `set_player_persona` call are both
found, ensure that the `set_player_name` call is processed first by moving it in front of the
`set_player_persona` call.
"""
set_player_name_call_exists = -1
set_player_persona_call_exists = -1
i = 0
for call in calls:
if "set_player_name" in call:
set_player_name_call_exists = i
elif "set_player_persona" in call:
set_player_persona_call_exists = i
i = i + 1
if set_player_name_call_exists > -1 and set_player_persona_call_exists > -1:
if set_player_name_call_exists > set_player_persona_call_exists:
calls.insert(set_player_persona_call_exists, calls.pop(set_player_name_call_exists))
TM.log.debug("SIMULATION SUITE: prepare calls - moved set_player_persona call", calls=calls)
return calls
def process_call(self, call:str) -> str:
"""
Processes a simulation call
Simulation alls are pseudo functions that are called by the simulation suite
We grab the function name by splitting against ( and taking the first element
if the SimulationSuite has a method with the name _call_{function_name} then we call it
if a function name could be found but we do not have a method to call we dont do anything
but we still return it as procssed as the AI can still interpret it as something later on
"""
if "(" not in call:
return None
function_name = call.split("(")[0]
if hasattr(self, f"call_{function_name}"):
TM.log.debug("SIMULATION SUITE CALL", call=call, function_name=function_name)
inject = f"The computer executes the function `{call}`"
return getattr(self, f"call_{function_name}")(call, inject)
return call
def call_set_simulation_goal(self, call:str, inject:str) -> str:
"""
Set's the simulation goal as a permanent pin
"""
TM.signals.status("busy", "Simulation suite setting goal.", as_scene_message=True)
TM.agents.world_state.save_world_entry(
entry_id="sim.goal",
text=self.player_message.raw,
meta={},
pin=True
)
TM.agents.director.log_action(
action=parse_sim_call_arguments(call),
action_description="The computer sets the goal for the simulation.",
)
return call
def call_change_environment(self, call:str, inject:str) -> str:
"""
Simulation changes the environment, this is entirely interpreted by the AI
and we dont need to do any logic on our end, so we just return the call
"""
TM.agents.director.log_action(
action=parse_sim_call_arguments(call),
action_description="The computer changes the environment of the simulation."
)
return call
def call_answer_question(self, call:str, inject:str) -> str:
"""
The player asked the simulation a query, we need to process this and have
the AI produce an answer
"""
TM.agents.narrator.action_to_narration(
action_name="progress_story",
narrative_direction=f"The computer calls the following function:\n\n{call}\n\nand answers the player's question.",
emit_message=True
)
def call_set_player_persona(self, call:str, inject:str) -> str:
"""
The simulation suite is altering the player persona
"""
player_character = TM.scene.get_player_character()
TM.signals.status("busy", "Simulation suite altering user persona.", as_scene_message=True)
character_attributes = TM.agents.world_state.extract_character_sheet(
name=player_character.name, text=inject, alteration_instructions=self.player_message.raw
)
TM.scene.set_character_attributes(player_character.name, character_attributes)
character_description = TM.agents.creator.determine_character_description(player_character.name)
TM.scene.set_character_description(player_character.name, character_description)
TM.log.debug("SIMULATION SUITE: transform player", attributes=character_attributes, description=character_description)
TM.agents.director.log_action(
action=parse_sim_call_arguments(call),
action_description="The computer transforms the player persona."
)
return call
def call_set_player_name(self, call:str, inject:str) -> str:
"""
The simulation suite is altering the player name
"""
player_character = TM.scene.get_player_character()
TM.signals.status("busy", "Simulation suite adjusting user identity.", as_scene_message=True)
character_name = TM.agents.creator.determine_character_name(instructions=f"{inject} - What is a fitting name for the player persona? Respond with the current name if it still fits.")
TM.log.debug("SIMULATION SUITE: player name", character_name=character_name)
if character_name != player_character.name:
TM.scene.set_character_name(player_character.name, character_name)
TM.agents.director.log_action(
action=parse_sim_call_arguments(call),
action_description=f"The computer changes the player's identity to {character_name}."
)
return call
def call_add_ai_character(self, call:str, inject:str) -> str:
# sometimes the AI will call this function an pass an inanimate object as the parameter
# we need to determine if this is the case and just ignore it
is_inanimate = TM.agents.world_state.answer_query_true_or_false(f"does the function `{call}` add an inanimate object, concept or abstract idea? (ANYTHING THAT IS NOT A CHARACTER THAT COULD BE PORTRAYED BY AN ACTOR)", call)
if is_inanimate:
TM.log.debug("SIMULATION SUITE: add npc - inanimate object / abstact idea - skipped", call=call)
return
# sometimes the AI will ask if the function adds a group of characters, we need to
# determine if this is the case
adds_group = TM.agents.world_state.answer_query_true_or_false(f"does the function `{call}` add MULTIPLE ai characters?", call)
TM.log.debug("SIMULATION SUITE: add npc", adds_group=adds_group)
TM.signals.status("busy", "Simulation suite adding character.", as_scene_message=True)
if not adds_group:
character_name = TM.agents.creator.determine_character_name(instructions=f"{inject} - what is the name of the character to be added to the scene? If no name can extracted from the text, extract a short descriptive name instead. Respond only with the name.")
else:
character_name = TM.agents.creator.determine_character_name(instructions=f"{inject} - what is the name of the group of characters to be added to the scene? If no name can extracted from the text, extract a short descriptive name instead. Respond only with the name.", group=True)
# sometimes add_ai_character and change_ai_character are called in the same instruction targeting
# the same character, if this happens we need to combine into a single add_ai_character call
has_change_ai_character_call = TM.agents.world_state.answer_query_true_or_false(f"Are there any calls to `change_ai_character` in the instruction for {character_name}?", "\n".join(self.calls))
if has_change_ai_character_call:
combined_arg = TM.prompt.request(
"combine-add-and-alter-ai-character",
dedupe_enabled=False,
calls="\n".join(self.calls),
character_name=character_name,
scene=TM.scene,
).replace("COMBINED ARGUMENT:", "").strip()
call = f"add_ai_character({combined_arg})"
inject = f"The computer executes the function `{call}`"
TM.signals.status("busy", f"Simulation suite adding character: {character_name}", as_scene_message=True)
TM.log.debug("SIMULATION SUITE: add npc", name=character_name)
npc = TM.agents.director.persist_character(character_name=character_name, content=self.player_message.raw+f"\n\n{inject}", determine_name=False)
self.added_npcs.append(npc.name)
TM.agents.world_state.add_detail_reinforcement(
character_name=npc.name,
detail="Goal",
instructions=f"Generate a goal for {npc.name}, based on the user's chosen simulation",
interval=25,
run_immediately=True
)
TM.log.debug("SIMULATION SUITE: added npc", npc=npc)
TM.agents.visual.generate_character_portrait(character_name=npc.name)
TM.agents.director.log_action(
action=parse_sim_call_arguments(call),
action_description=f"The computer adds {npc.name} to the simulation."
)
return call
####
def call_remove_ai_character(self, call:str, inject:str) -> str:
TM.signals.status("busy", "Simulation suite removing character.", as_scene_message=True)
character_name = TM.agents.creator.determine_character_name(instructions=f"{inject} - what is the name of the character being removed?", allowed_names=TM.scene.npc_character_names)
npc = TM.scene.get_character(character_name)
if npc:
TM.log.debug("SIMULATION SUITE: remove npc", npc=npc.name)
TM.agents.world_state.deactivate_character(action_name="deactivate_character", character_name=npc.name)
TM.agents.director.log_action(
action=parse_sim_call_arguments(call),
action_description=f"The computer removes {npc.name} from the simulation."
)
return call
def call_change_ai_character(self, call:str, inject:str) -> str:
TM.signals.status("busy", "Simulation suite altering character.", as_scene_message=True)
character_name = TM.agents.creator.determine_character_name(instructions=f"{inject} - what is the name of the character receiving the changes (before the change)?", allowed_names=TM.scene.npc_character_names)
if character_name in self.added_npcs:
# we dont want to change the character if it was just added
return
character_name_after = TM.agents.creator.determine_character_name(instructions=f"{inject} - what is the name of the character receiving the changes (after the changes)?")
npc = TM.scene.get_character(character_name)
if npc:
TM.signals.status("busy", f"Changing {character_name} -> {character_name_after}", as_scene_message=True)
TM.log.debug("SIMULATION SUITE: transform npc", npc=npc)
character_attributes = TM.agents.world_state.extract_character_sheet(
name=npc.name,
text=inject,
alteration_instructions=self.player_message.raw
)
TM.scene.set_character_attributes(npc.name, character_attributes)
character_description = TM.agents.creator.determine_character_description(npc.name)
TM.scene.set_character_description(npc.name, character_description)
TM.log.debug("SIMULATION SUITE: transform npc", attributes=character_attributes, description=character_description)
if character_name_after != character_name:
TM.scene.set_character_name(npc.name, character_name_after)
TM.agents.director.log_action(
action=parse_sim_call_arguments(call),
action_description=f"The computer transforms {npc.name}."
)
return call
def call_end_simulation(self, call:str, inject:str) -> str:
player_character = TM.scene.get_player_character()
explicit_command = TM.agents.world_state.answer_query_true_or_false("has the player explicitly asked to end the simulation?", self.player_message.raw)
if explicit_command:
TM.signals.status("busy", "Simulation suite ending current simulation.", as_scene_message=True)
TM.agents.narrator.action_to_narration(
action_name="progress_story",
narrative_direction=f"Narrate the computer ending the simulation, dissolving the environment and all artificial characters, erasing all memory of it and finally returning the player to the inactive simulation suite. List of artificial characters: {', '.join(TM.scene.npc_character_names)}. The player is also transformed back to their normal, non-descript persona as the form of {player_character.name} ceases to exist.",
emit_message=True
)
TM.scene.restore()
self.simulation_reset = True
TM.game_state.unset_var("instr.has_issued_instructions")
TM.game_state.unset_var("instr.lastprocessed_call")
TM.game_state.unset_var("instr.simulation_started")
TM.agents.director.log_action(
action=parse_sim_call_arguments(call),
action_description="The computer ends the simulation."
)
def finalize_round(self):
# track rounds
rounds = TM.game_state.get_var("instr.rounds", 0)
# increase rounds
TM.game_state.set_var("instr.rounds", rounds + 1, commit=False)
has_issued_instructions = TM.game_state.has_var("instr.has_issued_instructions")
if self.update_world_state:
self.run_update_world_state()
if self.player_message_is_instruction:
TM.scene.hide_message(self.player_message.id)
TM.game_state.set_var("instr.lastprocessed_call", self.player_message.id, commit=False)
TM.signals.status("success", MSG_PROCESSED_INSTRUCTIONS, as_scene_message=True)
elif self.player_message and not has_issued_instructions:
# simulation started, player message is NOT an instruction, and player has not given
# any instructions
self.guide_player()
elif self.player_message and not TM.scene.npc_character_names:
# simulation started, player message is NOT an instruction, but there are no npcs to interact with
self.narrate_round()
elif rounds % AUTO_NARRATE_INTERVAL == 0 and rounds and TM.scene.npc_character_names and has_issued_instructions:
# every N rounds, narrate the round
self.narrate_round()
def guide_player(self):
TM.agents.narrator.action_to_narration(
action_name="paraphrase",
narration=MSG_HELP,
emit_message=True
)
def narrate_round(self):
TM.agents.narrator.action_to_narration(
action_name="progress_story",
narrative_direction=PROMPT_NARRATE_ROUND,
emit_message=True
)
def run_update_world_state(self, force=False):
TM.log.debug("SIMULATION SUITE: update world state", force=force)
TM.signals.status("busy", "Simulation suite updating world state.", as_scene_message=True)
TM.agents.world_state.update_world_state(force=force)
TM.signals.status("success", "Simulation suite updated world state.", as_scene_message=True)
SimulationSuite().run()
def on_generation_cancelled(TM, exc):
"""
Called when user pressed the cancel button during the simulation suite
loop.
"""
TM.signals.status("success", "Simulation suite instructions cancelled", as_scene_message=True)
rounds = TM.game_state.get_var("instr.rounds", 0)
TM.log.debug("SIMULATION SUITE: command cancelled", rounds=rounds)