forked from art-from-the-machine/Mantella
-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathsettings_descriptions.json
144 lines (144 loc) · 12.6 KB
/
settings_descriptions.json
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
{
"Game": {
"game_id": "The game id of the game you are playing. Currently only 'skyrim', 'skyrimvr, 'fallout4' and 'fallout4vr' are supported.",
"conversation_manager_type": "The conversation manager to use for the game. Defaults to 'auto' to let game_config decide.",
"interface_type": "The interface type to use for the game. Defaults to 'auto' to let game_config decide.",
"behavior_manager": "The behavior manager to use for the game. Defaults to 'auto' to let game_config decide.",
},
"Language": {
"language": "The language all NPCs are told to speak in. Doesn't translate prompts or player responses.",
"end_conversation_keywords": "List of keywords that will end the conversation when spoken.",
"goodbye_npc_responses": "List of responses the NPC will give when the conversation ends.",
"collecting_thoughts_npc_responses": "The response the NPC will give when they are thinking."
},
"Microphone": {
"whisper_model": "The size of whisper model to use. E.g. 'tiny', 'tiny.en', 'base', 'base.en', 'small', 'small.en', 'medium', 'medium.en', 'large-v1', 'large-v2', 'large-v3', or 'large'. Or a path to a converted model directory, or a CTranslate2-converted Whisper model ID from the HF Hub. When a size or a model ID is configured, the converted model is downloaded from the Hugging Face Hub.",
"stt_language": "The language to use for speech-to-text.",
"stt_translate": "Whether to translate the speech-to-text.",
"whisper_process_device": "The whisper process device to use. Defaults to 'cpu', but can be set to 'cuda' to use the GPU.",
"whisper_type": "The whisper type to use. Defaults to 'faster_whisper', 'whisper_server' to use whisper_url's whisper server instead of the local one.",
"vad_filter": "Whether to use VAD filtering. Defaults to True.",
"beam_size": "The beam size for whisper. Defaults to 5.",
"whisper_compute_type": "The whisper compute type to use. Defaults to 'auto'.",
"whisper_cpu_threads": "The whisper CPU threads to use. Defaults to 4.",
"whisper_url": "The whisper URL to use. Defaults to 'http://127.0.0.1:8080/inference'.",
"audio_threshold": "The audio threshold. Defaults to 'auto'.",
"pause_threshold": "The pause threshold. Defaults to 0.5.",
"listen_timeout": "The timeout for listening."
},
"LanguageModel": {
"inference_engine": "The inference engine to use. Defaults to 'openai' to use OpenAI's API. Available: openai, llama-cpp-python, llava-cpp-python, transformers.",
"tokenizer_type": "The tokenizer type to use. Defaults to 'default' to use the default tokenizer for the inference engine. Only change this if you know why you need to change it.",
"prompt_style": "The prompt style to use. Defaults to 'default' to use the default prompt style for the inference engine. Only change this if you know why you need to change it.",
"maximum_local_tokens": "The maximum local tokens. Defaults to 4096.",
"max_response_sentences": "The maximum number of sentences in a response. Defaults to 3.",
"wait_time_buffer": "The wait time buffer. Defaults to 0.5.",
"stop": "A list of stop words to use. Defaults to ['<im_start>', '<im_end>'].",
"BOS_token": "The BOS token to use. Defaults to '<im_start>'.",
"EOS_token": "The EOS token to use. Defaults to '<im_end>'.",
"message_signifier": "The phrase that signifies a message. Defaults to '\\n'.",
"message_separator": "The phrase that separates messages. Defaults to '\\n'.",
"message_format": "The format of the message. Defaults to '[BOS_token][name][message_signifier][content][EOS_token][message_separator]'",
"min_conversation_length": "The minimum conversation length. Defaults to 5.",
"system_name": "The name of the sender of system messages. System messages are messages that are not from a character and are used to convey information to the LLM. Defaults to 'system'.",
"user_name": "The name of the default user. Defaults to 'user'. This is the name of the user that is used when the user is not specified. Chat completions(openai) will have this enabled for all user messages by default.",
"assistant_name": "The name of the assistant. Defaults to 'assistant'. This is the name of the assistant that is used when the assistant is not specified. Chat completions(openai) will have this enabled for all LLM messages by default.",
"assist_check": "Whether to check if the word 'assist' is in the messag and discard the message if it is. Defaults to False.",
"strip_smalls": "Whether to strip small messages. Defaults to False.",
"small_size": "The length of a small message. Defaults to 5.",
"same_output_limit": "Limits the number of times the same output can be repeated in a row. Defaults to 30.",
"conversation_limit_pct": "The percentage of context that can be filled before triggering a summarization. Defaults to 0.8.",
"reload_buffer": "The number of messages returned to the context after summarization. Defaults to 8."
},
"InferenceOptions": {
"temperature": "The temperature for inference. Higher values make the model more creative but less coherent, more spelling mistakes, and more hallucinations. Lower values make the model less creative but more coherent. Defaults to 0.7.",
"top_p": "Top p is used as a measure to filter out tokens that are less likely to occur. 0.9 means that the model will only consider tokens that are in the top 90% of likelihood. Defaults to 0.9. If you want to disable it, set it to 1.",
"min_p": "New sampling method that only considers tokens that are at least this likely to occur. Defaults to 0.05. Replaces top_p and top_k.",
"typical_p": "Typical p measures how similar the conditional probability of predicting a target token next is to the expected conditional probability of predicting a random token next, given the partial text already generated. If set to float < 1, the smallest set of the most locally typical tokens with probabilities that add up to typical_p or higher are kept for generation. Defaults to 1.0.",
"top_k": "Top k is used as a measure to filter out tokens that are less likely to occur. 40 means that the model will only consider the top 40 tokens. Very rudimentary, I don't recommend this over other sampling options if they're available. Defaults to 0. If you want to disable it, set it to 0.",
"repeat_penalty": "Penalizes tokens that are repeated in the output. Defaults to 1.2.",
"tfs_z": "Tail free sampling. https://www.trentonbricken.com/Tail-Free-Sampling/. Defaults to 1.0. If you want to disable it, set it to 1.0.",
"frequency_penalty": "Penalizes tokens that are appear frequently in the output. Defaults to 0.01.",
"presence_penalty": "Encourages the model to generate tokens that are not already in the output. Defaults to 0.0.",
"mirostat_mode": "Mirostat is a type of sampling that overrides everthing else but temperature. It will then change how it samples to force the model to match a certain perplexity target. Defaults to '0' to disable it. If you want to enable it, set it to '1' or '2' for Mode 1 or Mode 2.",
"mirostat_eta": "The mirostat eta, the learning rate for the perplexity target. Defaults to 0.1.",
"mirostat_tau": "The mirostat tau, the perplexity target. Defaults to 5.",
"max_tokens": "The maximum number of tokens to generate. Defaults to 512."
},
"Vision": {
"vision_enabled": "Whether vision is enabled. Defaults to True.",
"paddle_ocr": "Whether to use paddle OCR. Defaults to False.",
"ocr_lang": "The language to use for OCR. Defaults to 'en'.",
"ocr_use_angle_cls": "Whether to use angle classification for OCR. Defaults to False.",
"ocr_filter":"A list of strings to filter out from the OCR output.",
"append_system_image_near_end": "Whether to append the system image near the end of the message. Defaults to True.",
"image_message_depth": "The depth of the image message. Defaults to -1. -1 means right before the next message.",
"image_message": "The format of the message with the image.",
"ocr_resolution": "The resolution of the OCR ASCII block. Defaults to 256.",
"image_resolution": "The resolution of the clip model. Defaults to 672."
},
"openai_api": {
"llm": "The language model ID to use.",
"alternative_openai_api_base": "If you're not using OpenAI's API, you can set this to the base URL of the API you're using.",
"secret_key_file_path": "The path to the secret key file. This is where you store your OpenAI API key."
},
"llama_cpp_python": {
"model_path": "The path to the model file. e.g. 'mistral-7B.Q4_K_M.gguf'.",
"n_gpu_layers": "The number of layers to offload to the GPU. Defaults to 0.",
"n_threads": "The number of threads to use. Defaults to 4. Only use up to your physical core count. If you're using Intel, only use the number of your performance cores.",
"n_batch": "The size of the batch of tokenizes to send to the model at once. Defaults to 1.",
"tensor_split": "The ratio of how to split the layers across multiple GPUs. Defaults to []. [0.5,0.5] for 2 gpus split evenly, [0.3,0.7] for 2 gpus split unevenly.",
"main_gpu": "The ID of the main GPU to use. Defaults to 0 for GPU 1. If you're using multiple GPUs, set this to the ID of the GPU you want to use.",
"split_mode": "Whether to split the model across multiple GPUs. Defaults to 0. 0 = single gpu, 1 = split layers and kv across gpus, 2 = split rows across gpus.",
"use_mmap": "Whether to use mmap. Defaults to True.",
"use_mlock": "Whether to use mlock. Defaults to False.",
"n_threads_batch": "The number of threads to use for processing batches. Defaults to 1.",
"offload_kqv": "Whether to offload kqv. Defaults to True. Warning: Uses more VRAM."
},
"llava_cpp_python": {
"llava_clip_model_path": "The path to the clip model file. e.g. 'mmproj-model-f16.gguf', 'llava-clip.f16.gguf'.",
},
"transformers": {
"transformers_model_slug": "The HF indentifier for the model. e.g. 'mistralai/Mistral-7B-Instruct-v0.1'.",
"trust_remote_code": "Whether to trust remote code. Defaults to False.",
"device_map": "The device map. Defaults to 'cuda:0'.",
"load_in_8bit": "Whether to load in 8bit. Defaults to False."
},
"Speech": {
"tts_engine": "Either a string or a list of strings defining the TTS engine to use, or which engines to use in which fallback order. Defaults to 'xvasynth'.",
"end_conversation_wait_time": "The wait time after the conversation ends. Defaults to 1.",
"sentences_per_voiceline": "The number of sentences per voiceline generated. Defaults to 2."
},
"xVASynth": {
"xvasynth_path": "The path to the xVASynth executable directory.",
"xvasynth_process_device": "The device to use for xVASynth. Defaults to 'cpu'.",
"pace": "The pace of the voice. Defaults to 1.",
"use_cleanup": "Whether to use cleanup.",
"use_sr": "Whether to use sr.",
"xvasynth_base_url": "If you're using a remote xVASynth server, set this to the base URL of the server."
},
"xTTS": {
"xtts_server_folder": "The folder where the xTTS server is located.",
"xtts_base_url": "The base URL of the xTTS server.",
"xtts_data": "The sampling options for xTTS.",
"default_xtts_model": "The default xtts model. xTTS models can be used for several voices at once, or a single voice per model. If you want to use a single voice per model, add a model to the xtts models directory with the correct voice_model as it's name."
},
"Cleanup": {
"remove_mei_folders": "Whether to remove mei folders on startup."
},
"Debugging": {
"debug_mode": "Whether to enable debug mode. - Doesn't currently do anything.",
"play_audio_from_script": "Whether to play audio from script.",
"debug_character_name": "The debug character name.",
"debug_use_mic": "Whether to use the mic for debugging.",
"default_player_response": "The default player response.",
"debug_exit_on_first_exchange": "Whether to exit on the first exchange.",
"add_voicelines_to_all_voice_folders": "Whether to add voicelines to all voice folders."
},
"Config": {
"character_database_file": "The character database file path.",
"voice_model_ref_ids_file": "The voice model ref ids file path.",
"logging_file_path": "The logging file path.",
"port": "The port to use for the web configuration server."
}
}