""" Title: NovelAI Connector author: TheDimensionOfEternity author_url: https://github.com/TheDimensionofEternity licence: The Unlicense Make sure that the novelai-api module is installed in your environment before installing this function. it won't work otherwise. In modelfiles, you can specify the Assistant's name with the following syntax on the first line of the system prompt. Assistant=My Custom Model This is my first time publically submitting code and I haven't rigorously tested it, there are likely to be bugs. """ from pydantic import BaseModel, Field import asyncio from novelai_api import NovelAIAPI from fastapi import Request ## START OF BOILERPLATE import asyncio import json from datetime import datetime from logging import Logger, StreamHandler from os import environ as env from pathlib import Path from typing import Any, Optional, List, Dict, Callable, Union, Iterable, AsyncIterable from aiohttp import ClientSession from msgpackr.constants import UNDEFINED from novelai_api import NovelAIAPI, Keystore from novelai_api.utils import get_encryption_key from novelai_api.Msgpackr_Extensions import ( Ext20, Ext30, Ext31, Ext40, Ext41, Ext42, ) import novelai_api.utils import atexit class API: """ Boilerplate for the redundant parts. Using the object as a context manager will automatically login using the environment variables ``NAI_USERNAME`` and ``NAI_PASSWORD``. Usage: .. code-block:: python async with API() as api: api = api.api encryption_key = api.encryption_key logger = api.logger ... # Do stuff A custom base address can be passed to the constructor to replace the default (:attr:`BASE_ADDRESS <novelai_api.NovelAI_API.NovelAIAPI.BASE_ADDRESS>`) """ _username: str _password: str _token: str _session: ClientSession logger: Logger api: Optional[NovelAIAPI] def __init__( self, base_address: Optional[str] = None, username: str = "", password: str = "", token: str = "", ): dotenv = Path(".env") if dotenv.exists(): with dotenv.open("r") as f: for line in f: if "=" in line: key, value = line.strip().split("=", 1) env[key] = value.strip() if (username != "" or "NAI_USERNAME" in env) or ( password != "" or "NAI_PASSWORD" in env ): if not (username != "" or "NAI_USERNAME" in env): raise RuntimeError( "Please ensure that NAI_USERNAME is set in your environment" ) if not (password != "" or "NAI_PASSWORD" in env): raise RuntimeError( "Please ensure that NAI_PASSWORD is set in your environment" ) self._username = username if username != "" else env["NAI_USERNAME"] self._password = password if password != "" else env["NAI_PASSWORD"] self._using_token = False else: if not (token != "" or "NAI_TOKEN" in env): raise RuntimeError( "Please ensure that NAI_TOKEN is set in your environment" ) self._using_token = True self._token = token if isinstance(token, str) else env["NAI_TOKEN"] self.logger = Logger("NovelAI") self.logger.addHandler(StreamHandler()) self.api: NovelAIAPI = NovelAIAPI(logger=self.logger) if base_address is not None: self.api.BASE_ADDRESS = base_address atexit.register(self.on_exit) @property def encryption_key(self) -> bytes: if self._using_token: return bytes() return get_encryption_key(self._username, self._password) async def __aenter__(self): self._session = ClientSession() await self._session.__aenter__() self.api.attach_session(self._session) if self._using_token: await self.api.high_level.login_with_token(self._token) else: await self.api.high_level.login(self._username, self._password) return self def on_exit(self): asyncio.run(self.__aexit__()) async def __aexit__(self, exc_type=None, exc_val=None, exc_tb=None): await self._session.__aexit__(exc_type, exc_val, exc_tb) class JSONEncoder(json.JSONEncoder): """ Extended JSON encoder to support bytes """ def default(self, o: Any) -> Any: if isinstance(o, bytes): return o.hex() if o is UNDEFINED: return f"<UNDEFINED> {repr(o)}" if isinstance(o, datetime): return o.isoformat() return super().default(o) def dumps(e: Any) -> str: """ Shortcut to a configuration of json.dumps for consistency """ return json.dumps(e, indent=4, ensure_ascii=False, cls=JSONEncoder) def document_wrapper(func: Callable): def wrapper(*args, **kwargs): try: return func(*args, **kwargs) except TypeError: del novelai_api.utils.unpacker novelai_api.utils.unpacker = novelai_api.utils.Unpacker() novelai_api.utils.unpacker.register_extensions( Ext20, Ext30, Ext31, Ext40, Ext41, Ext42 ) return func(*args, **kwargs) return wrapper @document_wrapper def decrypt_user_data( items: List[Dict[str, Any]] | Dict[str, Any], keystore: Keystore, uncompress_document: bool = False, ) -> None: return novelai_api.utils.decrypt_user_data( items=items, keystore=keystore, uncompress_document=uncompress_document ) ### END OF BOILERPLATE from novelai_api.BanList import BanList from novelai_api.BiasGroup import BiasGroup from novelai_api.GlobalSettings import GlobalSettings from novelai_api.Preset import PREAMBLE, Model, Preset, Order from novelai_api.Tokenizer import Tokenizer from novelai_api.utils import b64_to_tokens class Pipe: class Valves(BaseModel): # NOVELAI_EMAIL: str = Field( # default="", # description="Email of account used to authenticate authenticating " # "requests to the NovelAI API.", # ) # NOVELAI_PASSWORD: str = Field( # default="", # description="Password of account used to authenticate authenticating " # "requests to the NovelAI API.", # ) NOVELAI_API_KEY: str = Field( default="", description="API key for authenticating requests to the NovelAI API.", ) USER_FORMATTING: str = Field( default="\n{user}: {data}\n", description="", ) ASSISTANT_FORMATTING: str = Field( default="\n{assistant}: {data}\n", description="", ) SYSTEM_FORMATTING: str = Field( default="\n {data}\n", description="", ) TOKEN_CUTOFF: str = Field( default="7900", description="The number of tokens to keep in history", ) def __init__(self): self.valves = self.Valves() def format_message( self, message: list, user: dict, assistant: str = "Assistant" ) -> str: formatted_message: str = "" additional_prompt = [] continue_message = {"role": "user", "content": "/continue"} should_continue: bool = message[-1]["content"].strip() == "/continue" while continue_message in message: message.remove(continue_message) if message[-1]["role"] != "assistant": additional_prompt = [{"role": "assistant", "content": ""}] for item in message + additional_prompt: content = item["content"] match item["role"]: case "user": formatted_message += self.valves.USER_FORMATTING.format( user=user["name"].title(), assistant=assistant.title(), data=content, ) case "assistant": formatted_message += self.valves.ASSISTANT_FORMATTING.format( user=user["name"].title(), assistant=assistant.title(), data=content, ) case "system": formatted_message += self.valves.SYSTEM_FORMATTING.format( user=user["name"].title(), assistant=assistant.title(), data=content.format( user=user["name"].title(), assistant=assistant, ), ) case _: raise ValueError("Uhh, how'd this happen? invalid role.") formatted_message = formatted_message.strip() if should_continue: formatted_message += "\n" return formatted_message def pipes(self): # if (not self.valves.NOVELAI_EMAIL and not self.valves.NOVELAI_PASSWORD if not self.valves.NOVELAI_API_KEY: return [ {"id": "error", "name": "Proper NovelAI Authentication not provided."}, ] erato_models = [ { "id": f"Erato/{preset_name.replace(' ','-')}", "name": f"{preset_name} (Erato)", } for preset_name in self.ERATO_PRESETS ] kayra_models = [ { "id": f"Kayra/{preset_name.replace(' ','-')}", "name": f"{preset_name} (Kayra)", } for preset_name in self.KAYRA_PRESETS ] clio_models = [ { "id": f"Clio/{preset_name.replace(' ','-')}", "name": f"{preset_name} (Clio)", } for preset_name in self.CLIO_PRESETS ] models = erato_models + kayra_models + clio_models return models[:] @staticmethod async def generate_stream( api: NovelAIAPI, prompt: Union[List[int], str], model: Model, preset: Preset, global_settings: GlobalSettings, bad_words: Optional[Union[Iterable[BanList], BanList]] = None, biases: Optional[Union[Iterable[BiasGroup], BiasGroup]] = None, prefix: Optional[str] = None, stop_sequences: Optional[Union[List[int], str]] = None, **kwargs, ) -> AsyncIterable[Dict[str, Any]]: bytes_per_token: int = 2 if model == Model.Erato: bytes_per_token = 4 prev_tokens = [] async for e in api.high_level._generate( prompt, model, preset, global_settings, bad_words, biases=biases, prefix=prefix, stop_sequences=stop_sequences, stream=True, **kwargs, ): def isMatch(token) -> int: nonlocal prev_tokens is_match: int = 0 for sequence in stop_sequences: sequence_additional = [sequence[0]] + [34184] + sequence[1:] if prev_tokens == []: break elif prev_tokens == sequence: is_match = 2 elif prev_tokens == sequence[: len(prev_tokens)]: is_match = 1 break elif prev_tokens == sequence_additional[: len(prev_tokens)]: del prev_tokens[1] is_match = 1 break return is_match def do_work(token): nonlocal prev_tokens output_data = "" prev_tokens.extend(token) is_match = isMatch(token) if not is_match: output_data += Tokenizer.decode(model, prev_tokens) prev_tokens = [] prev_tokens.extend(token) if not isMatch(token): prev_tokens = [] elif is_match == 2: output_data += Tokenizer.decode( model, [prev_tokens[0]] ).removesuffix("\n") return output_data token = b64_to_tokens(json.loads(e)["token"], bytes_per_token) yield do_work(token) @classmethod def generate_stop_sequences(cls, model: Model, *roles: str) -> List[List[int]]: stop_sequences: List[str] = [] for role in roles: for sequence in cls.STOP_SEQUENCES: stop_sequences.append(sequence.format(role=role.title())) stop_sequences += cls.ADDITIONAL_STOP_SEQUENCES tokenized_stop_sequences: List[List[int]] = [] for sequence in stop_sequences: tokens = Tokenizer.encode(model=model, o=sequence) if model == Model.Erato: tokens = tokens[1:] tokenized_stop_sequences.append(tokens) return tokenized_stop_sequences async def pipe(self, body: dict, __user__: dict): MESSAGES: List[Dict[str:str]] = body["messages"] model: str = body["model"].split(".", 1)[-1] STREAM: bool = body["stream"] assistant = "Assistant" if str(MESSAGES[0]["content"]).split("\n", 1)[0].startswith("Assistant="): assistant = MESSAGES[0]["content"].split("\n", 1)[0].split("=", 1)[-1] MESSAGES[0]["content"] = MESSAGES[0]["content"].split("\n", 1)[1] async with API( username="", password="", token=self.valves.NOVELAI_API_KEY, ) as api_handler: api = api_handler.api logger = api_handler.logger model_name: str = model.split("/")[0] model_preset: str = model.split("/")[1] match model_name: case "Erato": preset_gen: dict = self.PRESETS[ "Erato-" + model_preset.replace(" ", "-") ] model = Model.Erato case "Kayra": preset_gen: dict = self.PRESETS[ model_preset.replace(" ", "-") + "-Kayra" ] model = Model.Kayra case "Clio": preset_gen: dict = self.PRESETS[ "Clio-" + model_preset.replace(" ", "-") ] model = Model.Clio case _: yield "error" preset_gen["order"] = [self.INT_TO_ORDER[o] for o in preset_gen["order"]] prompt: str = PREAMBLE[model] + self.format_message( MESSAGES, __user__, assistant ) token_prompt = Tokenizer.encode(model=model, o=prompt) token_prompt.reverse() token_prompt = token_prompt[: int(self.valves.TOKEN_CUTOFF)] token_prompt.reverse() prompt = Tokenizer.decode(model=model, o=token_prompt) del token_prompt preset = Preset( model_preset, model, ) for key, value in preset_gen.items(): if hasattr(preset, key): setattr(preset, key, value) if key == "order": preset.sampling_options = [True] * len(value) global_settings = GlobalSettings(num_logprobs=GlobalSettings.NO_LOGPROBS) global_settings.bias_dinkus_asterism = True global_settings.rep_pen_whitelist = True global_settings.generate_until_sentence = True bad_words: Optional[BanList] = None if bad_words is not None: bad_words.add( "<|startoftext|>", "<|endoftext|>", "<||ENDOFFTEXT||>", "<|endoftext|>", ) bias_groups: List[BiasGroup] = [] module = None stop_sequence = self.generate_stop_sequences( model, "***", __user__["name"] + ":", f"{assistant}:", ) bytes_per_token = 2 if model == Model.Erato: bytes_per_token = 4 if not STREAM: gen = await api.high_level.generate( prompt, model, preset, global_settings, bad_words=bad_words, biases=bias_groups, prefix=module, stop_sequences=stop_sequence, ) yield Tokenizer.decode( model, b64_to_tokens(gen["output"], bytes_per_token) ) else: async for e in self.generate_stream( api, prompt, model, preset, global_settings, bad_words, biases=bias_groups, prefix=module, stop_sequences=stop_sequence, ): yield e INT_TO_ORDER = { 0: Order.Temperature, 1: Order.Top_K, 2: Order.Top_P, 3: Order.TFS, 4: Order.Top_A, 5: Order.Typical_P, 8: Order.Mirostat, 9: Order.Unified, 10: Order.Min_p, } ERATO_PRESETS = [ "Shosetsu", "Wilder", "Dragonfruit", "Zany Scribe", "Golden Arrow", ] KAYRA_PRESETS = [ "Fresh Coffee", "Asper", "Writers Daemon", "Carefree", "Stelenes", ] """OLD_KAYRA_PRESETS = [ "Tesseract", "Blended Coffee", "CosmicCube", "Tea Time", "Pilotfish", "Green Active Writer", "Blook", "Pro Writer", ]""" CLIO_PRESETS = [ "Vingt-Un", "Long Press", "Edgewise", "Fresh Coffee", "Talker Chat", ] """OLD_CLIO_PRESETS = [ "Keelback", ]""" STOP_SEQUENCES = [ "\n{role}", ".\n{role}", "!\n{role}", "?\n{role}", "*\n{role}", '"\n{role}', "_\n{role}", "...\n{role}", '."\n{role}', '?"\n{role}', '!"\n{role}', ".*\n{role}", ")\n{role}", ".)\n{role}", ] ADDITIONAL_STOP_SEQUENCES = ["\n***"] PRESETS = { "Fresh-Coffee-Kayra": { "order": [0, 1, 2, 3], "temperature": 1, "max_length": 150, "min_length": 1, "top_k": 25, "top_p": 1, "tail_free_sampling": 0.925, "repetition_penalty": 1.9, "repetition_penalty_range": 768, "repetition_penalty_slope": 1, "repetition_penalty_frequency": 0.0025, "repetition_penalty_presence": 0.001, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "off", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Asper-Kayra": { "order": [5, 0, 1, 3], "temperature": 1.16, "max_length": 150, "min_length": 1, "top_k": 175, "typical_p": 0.96, "tail_free_sampling": 0.994, "repetition_penalty": 1.68, "repetition_penalty_range": 2240, "repetition_penalty_slope": 1.5, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0.005, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "medium", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Erato-Wilder": { "max_context": 8000, "temperature": 1, "max_length": 150, "min_length": 1, "top_k": 300, "top_p": 0.98, "top_a": 0.004, "typical_p": 0.96, "tail_free_sampling": 0.96, "repetition_penalty": 1.48, "repetition_penalty_range": 2240, "repetition_penalty_slope": 0.64, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "phrase_rep_pen": "medium", "mirostat_lr": 1, "mirostat_tau": 0, "math1_temp": -0.0485, "math1_quad": 0.145, "math1_quad_entropy_scale": 0, "min_p": 0.02, "order": [9, 10], }, "Erato-Dragonfruit": { "max_context": 8000, "temperature": 1.37, "max_length": 150, "min_length": 1, "top_k": 0, "top_p": 1, "top_a": 0.1, "typical_p": 0.875, "tail_free_sampling": 0.87, "repetition_penalty": 3.25, "repetition_penalty_range": 6000, "repetition_penalty_slope": 3.25, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "phrase_rep_pen": "off", "mirostat_lr": 0.2, "mirostat_tau": 4, "math1_temp": 0.9, "math1_quad": 0.07, "math1_quad_entropy_scale": -0.05, "min_p": 0.035, "order": [0, 5, 9, 10, 8, 4], }, "Edgewise-Clio": { "order": [4, 0, 5, 3, 2], "temperature": 1.09, "max_length": 150, "min_length": 1, "top_p": 0.969, "top_a": 0.09, "typical_p": 0.99, "tail_free_sampling": 0.969, "repetition_penalty": 1.09, "repetition_penalty_range": 8192, "repetition_penalty_slope": 0.069, "repetition_penalty_frequency": 0.006, "repetition_penalty_presence": 0.009, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "very_light", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Vingt-Un-Clio": { "order": [0, 5, 3, 2, 1], "temperature": 1.21, "max_length": 40, "min_length": 1, "top_k": 0, "top_p": 0.912, "top_a": 1, "typical_p": 0.912, "tail_free_sampling": 0.921, "repetition_penalty": 1.21, "repetition_penalty_range": 321, "repetition_penalty_slope": 3.33, "repetition_penalty_frequency": 0.00621, "repetition_penalty_presence": 0, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "very_light", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Fresh-Coffee-Clio": { "order": [0, 1, 2, 3], "temperature": 1, "max_length": 40, "min_length": 1, "top_k": 25, "top_p": 1, "top_a": 0, "typical_p": 1, "tail_free_sampling": 0.925, "repetition_penalty": 1.9, "repetition_penalty_range": 768, "repetition_penalty_slope": 3.33, "repetition_penalty_frequency": 0.0025, "repetition_penalty_presence": 0.001, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "very_light", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Writers-Daemon-Kayra": { "order": [8, 0, 5, 3, 2, 4], "temperature": 1.5, "max_length": 150, "min_length": 1, "top_a": 0.02, "top_p": 0.95, "typical_p": 0.95, "tail_free_sampling": 0.95, "mirostat_lr": 0.25, "mirostat_tau": 5, "repetition_penalty": 1.625, "repetition_penalty_range": 2016, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "very_aggressive", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Talker-Chat-Clio": { "order": [1, 5, 0, 2, 3, 4], "temperature": 1.5, "max_length": 150, "min_length": 1, "top_k": 10, "top_p": 0.75, "top_a": 0.08, "typical_p": 0.975, "tail_free_sampling": 0.967, "repetition_penalty": 2.25, "repetition_penalty_range": 8192, "repetition_penalty_slope": 0.09, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0.005, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "very_light", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Tesseract-Kayra": { "order": [0, 5], "temperature": 0.895, "max_length": 150, "min_length": 1, "typical_p": 0.9, "repetition_penalty": 2, "repetition_penalty_slope": 3.2, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "repetition_penalty_range": 4048, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "aggressive", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Long-Press-Clio": { "order": [0, 4, 1, 5, 3], "temperature": 1.155, "max_length": 40, "min_length": 1, "top_k": 25, "top_a": 0.3, "top_p": 1, "typical_p": 0.96, "tail_free_sampling": 0.895, "repetition_penalty": 1.0125, "repetition_penalty_range": 2048, "repetition_penalty_slope": 3.33, "repetition_penalty_frequency": 0.011, "repetition_penalty_presence": 0.005, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "very_light", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Keelback-Clio": { "order": [4, 5, 0, 3], "temperature": 1.18, "max_length": 40, "min_length": 1, "top_a": 0.022, "top_k": 0, "top_p": 1, "typical_p": 0.9, "tail_free_sampling": 0.956, "repetition_penalty": 1.25, "repetition_penalty_range": 4096, "repetition_penalty_slope": 0.9, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "very_light", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Blended-Coffee-Kayra": { "order": [0, 1, 2, 3], "temperature": 1, "max_length": 150, "min_length": 1, "top_k": 25, "top_p": 1, "tail_free_sampling": 0.925, "repetition_penalty": 1.6, "repetition_penalty_frequency": 0.001, "repetition_penalty_range": 0, "repetition_penalty_presence": 0, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "medium", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "CosmicCube-Kayra": { "order": [8, 5, 0, 3], "temperature": 0.9, "max_length": 150, "min_length": 1, "typical_p": 0.95, "tail_free_sampling": 0.92, "mirostat_lr": 0.22, "mirostat_tau": 4.95, "repetition_penalty": 3, "repetition_penalty_range": 4000, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "off", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Tea-Time-Kayra": { "order": [5, 0, 4], "temperature": 1, "max_length": 150, "min_length": 1, "top_a": 0.017, "typical_p": 0.975, "repetition_penalty": 3, "repetition_penalty_slope": 0.09, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "repetition_penalty_range": 7680, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "aggressive", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Pilotfish-Kayra": { "order": [0, 4, 1, 2, 5, 3], "temperature": 1.31, "max_length": 150, "min_length": 1, "top_k": 25, "top_p": 0.97, "top_a": 0.18, "typical_p": 0.98, "tail_free_sampling": 1, "repetition_penalty": 1.55, "repetition_penalty_frequency": 0.00075, "repetition_penalty_presence": 0.00085, "repetition_penalty_range": 8192, "repetition_penalty_slope": 1.8, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "medium", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Green-Active-Writer-Kayra": { "order": [0, 8, 5, 3], "temperature": 1.5, "max_length": 150, "min_length": 1, "typical_p": 0.95, "tail_free_sampling": 0.95, "mirostat_lr": 0.2, "mirostat_tau": 5.5, "repetition_penalty": 1, "repetition_penalty_range": 1632, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "very_aggressive", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Blook-Kayra": { "order": [2, 3, 1, 0], "temperature": 1, "max_length": 150, "min_length": 1, "top_k": 0, "top_p": 0.96, "tail_free_sampling": 0.96, "repetition_penalty": 2, "repetition_penalty_slope": 1, "repetition_penalty_frequency": 0.02, "repetition_penalty_range": 0, "repetition_penalty_presence": 0.3, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "very_aggressive", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Pro-Writer-Kayra": { "order": [3, 4, 5, 0], "temperature": 1.06, "max_length": 150, "min_length": 1, "top_a": 0.146, "typical_p": 0.976, "tail_free_sampling": 0.969, "repetition_penalty": 1.86, "repetition_penalty_slope": 2.33, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "repetition_penalty_range": 2048, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "medium", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Erato-Shosetsu": { "max_context": 8000, "temperature": 1, "max_length": 150, "min_length": 1, "top_k": 50, "top_p": 0.85, "top_a": 1, "typical_p": 1, "tail_free_sampling": 0.895, "repetition_penalty": 1.63, "repetition_penalty_range": 1024, "repetition_penalty_slope": 3.33, "repetition_penalty_frequency": 0.0035, "repetition_penalty_presence": 0, "phrase_rep_pen": "medium", "mirostat_lr": 1, "mirostat_tau": 0, "math1_temp": 0.3, "math1_quad": 0.0645, "math1_quad_entropy_scale": 0.05, "min_p": 0.05, "order": [9, 10], }, "Carefree-Kayra": { "order": [2, 3, 0, 4, 1], "temperature": 1.35, "max_length": 150, "min_length": 1, "top_k": 15, "top_p": 0.85, "top_a": 0.1, "tail_free_sampling": 0.915, "repetition_penalty": 2.8, "repetition_penalty_range": 2048, "repetition_penalty_slope": 0.02, "repetition_penalty_frequency": 0.02, "repetition_penalty_presence": 0, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "aggressive", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Erato-Zany-Scribe": { "max_context": 8000, "temperature": 1, "max_length": 150, "min_length": 1, "top_k": 0, "top_p": 0.99, "top_a": 1, "typical_p": 1, "tail_free_sampling": 0.99, "repetition_penalty": 1, "repetition_penalty_range": 64, "repetition_penalty_slope": 1, "repetition_penalty_frequency": 0.75, "repetition_penalty_presence": 1.5, "phrase_rep_pen": "medium", "mirostat_lr": 1, "mirostat_tau": 1, "math1_temp": -0.4, "math1_quad": 0.6, "math1_quad_entropy_scale": -0.1, "min_p": 0.08, "order": [9, 2], }, "Stelenes-Kayra": { "order": [3, 0, 5], "temperature": 2.5, "max_length": 150, "min_length": 1, "typical_p": 0.969, "tail_free_sampling": 0.941, "repetition_penalty": 1, "repetition_penalty_range": 1024, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "use_cache": False, "return_full_text": False, "prefix": "vanilla", "phrase_rep_pen": "medium", "max_context": 7800, "min_p": 0, "math1_temp": 1, "math1_quad": 0, "math1_quad_entropy_scale": 0, }, "Erato-Golden-Arrow": { "max_context": 8000, "temperature": 1, "max_length": 150, "min_length": 1, "top_k": 0, "top_p": 0.995, "top_a": 1, "typical_p": 1, "tail_free_sampling": 0.87, "repetition_penalty": 1.5, "repetition_penalty_range": 2240, "repetition_penalty_slope": 1, "repetition_penalty_frequency": 0, "repetition_penalty_presence": 0, "phrase_rep_pen": "light", "mirostat_lr": 1, "mirostat_tau": 0, "math1_temp": 0.3, "math1_quad": 0.19, "math1_quad_entropy_scale": 0, "min_p": 0, "order": [9, 2], }, }