src/enums.py

from enum import Enum


class PromptType(Enum):
    template = -3
    unknown = -2
    custom = -1
    plain = 0
    instruct = 1
    quality = 2
    human_bot = 3
    dai_faq = 4
    summarize = 5
    simple_instruct = 6
    instruct_vicuna = 7
    instruct_with_end = 8
    human_bot_orig = 9
    prompt_answer = 10
    open_assistant = 11
    wizard_lm = 12
    wizard_mega = 13
    instruct_vicuna2 = 14
    instruct_vicuna3 = 15
    wizard2 = 16
    wizard3 = 17
    instruct_simple = 18
    wizard_vicuna = 19
    openai = 20
    openai_chat = 21
    gptj = 22
    prompt_answer_openllama = 23
    vicuna11 = 24
    mptinstruct = 25
    mptchat = 26
    falcon = 27
    guanaco = 28
    llama2 = 29
    beluga = 30
    wizard3nospace = 31
    one_shot = 32
    falcon_chat = 33
    mistral = 34
    zephyr = 35
    xwin = 36
    mistrallite = 37
    aquila = 38
    aquila_simple = 39
    aquila_legacy = 40
    aquila_v1 = 41
    mistralgerman = 42
    deepseek_coder = 43
    open_chat = 44
    open_chat_correct = 45
    open_chat_code = 46
    anthropic = 47
    orca2 = 48
    jais = 49
    yi = 50
    xwincoder = 51
    xwinmath = 52
    vicuna11nosys = 53
    zephyr0 = 54
    google = 55
    docsgpt = 56
    open_chat_math = 57
    mistralai = 58
    mixtral = 59
    mixtralnosys = 60
    orion = 61
    sciphi = 62
    beacon = 63
    beacon2 = 64
    llava = 65
    danube = 66
    gemma = 67
    qwen = 68
    sealion = 69
    groq = 70
    aya = 71
    idefics2 = 72


class DocumentSubset(Enum):
    Relevant = 0
    RelSources = 1
    TopKSources = 2


non_query_commands = [
    DocumentSubset.RelSources.name,
    DocumentSubset.TopKSources.name
]


class DocumentChoice(Enum):
    ALL = 'All'


class LangChainMode(Enum):
    """LangChain mode"""

    DISABLED = "Disabled"
    LLM = "LLM"
    WIKI = "wiki"
    WIKI_FULL = "wiki_full"
    USER_DATA = "UserData"
    MY_DATA = "MyData"
    GITHUB_H2OGPT = "github h2oGPT"
    H2O_DAI_DOCS = "DriverlessAI docs"


class LangChainTypes(Enum):
    SHARED = 'shared'
    PERSONAL = 'personal'
    EITHER = 'either'  # used when user did not pass which one, so need to try both


# modes should not be removed from visible list or added by name
langchain_modes_intrinsic = [LangChainMode.DISABLED.value,
                             LangChainMode.LLM.value,
                             LangChainMode.MY_DATA.value]

langchain_modes_non_db = [LangChainMode.DISABLED.value,
                          LangChainMode.LLM.value]


class LangChainAction(Enum):
    """LangChain action"""

    QUERY = "Query"
    # WIP:
    # SUMMARIZE_MAP = "Summarize_map_reduce"
    SUMMARIZE_MAP = "Summarize"
    SUMMARIZE_ALL = "Summarize_all"
    SUMMARIZE_REFINE = "Summarize_refine"
    EXTRACT = "Extract"
    IMAGE_GENERATE = "ImageGen"
    IMAGE_CHANGE = "ImageChange"
    IMAGE_QUERY = "ImageQuery"
    IMAGE_STYLE = "ImageStyle"


valid_imagegen_models = ['sdxl_turbo', 'sdxl', 'sd3', 'playv2', 'flux.1-dev', 'flux.1-schnell']
valid_imagechange_models = ['sdxl_change']
valid_imagestyle_models = ['sdxl_style']

# rest are not implemented fully
base_langchain_actions = [LangChainAction.QUERY.value, LangChainAction.SUMMARIZE_MAP.value,
                          LangChainAction.EXTRACT.value,
                          LangChainAction.IMAGE_GENERATE.value,
                          LangChainAction.IMAGE_CHANGE.value,
                          LangChainAction.IMAGE_QUERY.value,
                          ]


class LangChainAgent(Enum):
    """LangChain agents"""

    SEARCH = "Search"
    COLLECTION = "Collection"
    PYTHON = "Python"
    CSV = "CSV"
    PANDAS = "Pandas"
    JSON = 'JSON'
    SMART = 'SMART'
    AUTOGPT = 'AUTOGPT'
    AIBEN = 'AIBEN'


no_server_str = no_lora_str = no_model_str = '[]'

# from:
# /home/jon/miniconda3/envs/h2ogpt/lib/python3.10/site-packages/langchain_community/llms/openai.py
# but needed since ChatOpenAI doesn't have this information
gpt_token_mapping = {
    "gpt-4": 8192,
    "gpt-4-0314": 8192,
    "gpt-4-0613": 8192,  # supports function tools
    "gpt-4-32k": 32768,
    "gpt-4-32k-0314": 32768,
    "gpt-4-32k-0613": 32768,  # supports function tools
    "gpt-3.5-turbo": 4096,
    "gpt-3.5-turbo-0301": 4096,
    "gpt-3.5-turbo-0613": 4096,  # supports function tools
    "gpt-3.5-turbo-16k": 16385,
    "gpt-3.5-turbo-16k-0613": 16385,  # supports function tools
    "gpt-3.5-turbo-instruct": 4096,
    "gpt-4-1106-preview": 128000,  # 4096 output
    "gpt-35-turbo-1106": 16385,  # 4096 output
    "gpt-4-vision-preview": 128000,  # 4096 output
    "gpt-4-1106-vision-preview": 128000,  # 4096 output
    "gpt-4-turbo-2024-04-09": 128000,  # 4096 output
    "gpt-4o": 128000,  # 4096 output
    "gpt-4o-2024-05-13": 128000,  # 4096 output
    "gpt-4o-2024-08-06": 128000,  # 4096 output
    "gpt-4o-mini": 128000,  # 16384 output
    # leave room for reasoning tokens
    "o1-preview": 128000,  # 4096 output
    "o1-mini": 128000,  # 4096 output
}
model_token_mapping = gpt_token_mapping.copy()
model_token_mapping.update({
    "text-ada-001": 2049,
    "ada": 2049,
    "text-babbage-001": 2040,
    "babbage": 2049,
    "text-curie-001": 2049,
    "curie": 2049,
    "davinci": 2049,
    "text-davinci-003": 4097,
    "text-davinci-002": 4097,
    "code-davinci-002": 8001,
    "code-davinci-001": 8001,
    "code-cushman-002": 2048,
    "code-cushman-001": 2048,
})

anthropic_mapping = {
    "claude-2.1": 200000,
    "claude-2": 100000,
    "claude-2.0": 100000,
    "claude-instant-1.2": 100000,
    "claude-3-opus-20240229": 200000,
    "claude-3-sonnet-20240229": 200000,
    "claude-3-5-sonnet-20241022": 200000,
    "claude-3-5-sonnet-latest": 200000,
    "claude-3-5-sonnet-20240620": 200000,
    "claude-3-haiku-20240307": 200000,
    "claude-3-5-haiku-20241022": 200000,
}

anthropic_mapping_outputs = {
    "claude-2.1": 4096,
    "claude-2": 4096,
    "claude-2.0": 4096,
    "claude-instant-1.2": 4096,
    "claude-3-opus-20240229": 4096,
    "claude-3-sonnet-20240229": 4096,
    "claude-3-5-sonnet-20240620": 8192,
    "claude-3-5-sonnet-20241022": 8192,
    "claude-3-5-sonnet-latest": 8192,
    "claude-3-haiku-20240307": 4096,
    "claude-3-5-haiku-20241022": 8192,
}

anthropic_prompt_caching = ["claude-3-opus-20240229",
                            "claude-3-5-sonnet-20241022",
                            "claude-3-5-sonnet-latest",
                            "claude-3-5-sonnet-20240620",
                            "claude-3-haiku-20240307",
                            "claude-3-5-haiku-20241022",
                            ]

claude3imagetag = 'claude-3-image'
gpt4imagetag = 'gpt-4-image'
geminiimagetag = 'gemini-image'
gemini15imagetag = 'gemini15-image'

claude3_image_tokens = 1334
gemini_image_tokens = 5000
gpt4_image_tokens = 1000

llava16_image_tokens = 2880
llava16_model_max_length = 4096
llava16_image_fudge = 50

# https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/gemini
#  Invalid argument provided to Gemini: 400 Please use fewer than 16 images in your request to models/gemini-pro-vision
# 4MB *total* limit of any prompt.  But only supports 16 images when doing fileData, needs to point to some gcp location
geminiimage_num_max = 15
# For gemini-1.5-pro, you can specify any combination and number of text, image, video, and audio files. The token limit is 1,000,000.
# no real limit, just set at 30
gemini15image_num_max = 30

# https://docs.anthropic.com/claude/docs/vision#image-best-practices
# https://github.com/anthropics/anthropic-cookbook/blob/main/multimodal/reading_charts_graphs_powerpoints.ipynb
# 5MB per image
claude3image_num_max = 20
# much worse image handling for many images.  Even 3 images gets confused.
claude3_haiku_image_num_max = 20
# https://platform.openai.com/docs/guides/vision
# 20MB per image request (they say per image but that's wrong)
# gpt-4o: ValueError: Error code: 400 - {'error': {'code': 'BadRequest', 'message': 'Too many images in request. Max is 10.', 'param': None, 'type': None}}
gpt4image_num_max = 10
# gpt-4o: ValueError: Error code: 400 - {'error': {'code': 'BadRequest', 'message': 'Too many images in request. Max is 20.', 'param': None, 'type': None}}
gpt4turbo_image_num_max = 20

# can be any number, but queued after --limit-model-concurrency <number> for some <number> e.g. 5
llava_num_max = 10

# really just limited by GPU memory, beyond 5 fails for single 80GB H100 or up to 8 images works for 2*80GB H100 before tokens run out for 1kx1k images
# but they don't do good with multiple images, so rely upon batching and pass -2 for model_lock value or CLI value
internvl_num_max = 5
internvl2_num_max = 10

images_num_max_dict = {'gpt-4-vision-preview': gpt4image_num_max,
                       'gpt-4-turbo-2024-04-09': gpt4turbo_image_num_max,
                       'gpt-4o': gpt4turbo_image_num_max,
                       'gpt-4o-2024-05-13': gpt4turbo_image_num_max,
                       'gpt-4o-2024-08-06': gpt4turbo_image_num_max,
                       'gpt-4o-mini': gpt4turbo_image_num_max,
                       'gpt-4o-mini-2024-07-18': gpt4turbo_image_num_max,
                       'gemini-pro-vision': geminiimage_num_max,
                       'gemini-1.5-pro-latest': gemini15image_num_max,
                       'gemini-1.5-flash-latest': gemini15image_num_max,
                       'claude-3-opus-20240229': claude3image_num_max,
                       'claude-3-sonnet-20240229': claude3image_num_max,
                       'claude-3-5-sonnet-20240620': claude3image_num_max,
                       'claude-3-5-sonnet-20241022': claude3image_num_max,
                       'claude-3-5-sonnet-latest': claude3image_num_max,
                       'claude-3-haiku-20240307': claude3_haiku_image_num_max,
                       'claude-3-5-haiku-20241022': claude3_haiku_image_num_max,
                       'liuhaotian/llava-v1.6-34b': 1,  # for lmdeploy
                       'liuhaotian/llava-v1.6-vicuna-13b': 1,  # for lmdeploy
                       'HuggingFaceM4/idefics2-8b-chatty': 10,
                       'lmms-lab/llama3-llava-next-8b': 2,
                       'OpenGVLab/InternVL-Chat-V1-5': internvl_num_max,
                       'THUDM/cogvlm2-llama3-chat-19B': 2,
                       'microsoft/Phi-3-vision-128k-instruct': 1,  # only 1 possible with vllm
                       }
for model_name in ["OpenGVLab/InternVL2-1B", "OpenGVLab/InternVL2-2B", "OpenGVLab/InternVL2-4B",
                   "OpenGVLab/InternVL2-8B", "OpenGVLab/InternVL2-26B", "OpenGVLab/InternVL2-40"]:
    images_num_max_dict[model_name] = internvl2_num_max

# llava34b sometimes runs out of tokens and finishes due to token limits, let's restrict
images_limit_max_new_tokens_list = ['liuhaotian/llava-v1.6-vicuna-13b', 'liuhaotian/llava-v1.6-34b']
images_limit_max_new_tokens = 512

# https://ai.google.dev/models/gemini
# gemini-1.0-pro
google_mapping = {
    "gemini-pro": 30720,
    "gemini-1.0-pro-latest": 30720,
    "gemini-pro-vision": 12288,
    "gemini-1.0-pro-vision-latest": 12288,
    "gemini-1.0-ultra-latest": 30720,
    "gemini-ultra": 30720,
    "gemini-1.5-pro-latest": 1048576,
    "gemini-1.5-flash-latest": 1048576,
}

# FIXME: at least via current API:
google_mapping_outputs = {
    "gemini-pro": 2048,
    "gemini-1.0-pro-latest": 2048,
    "gemini-pro-vision": 4096,
    "gemini-1.0-pro-vision-latest": 4096,
    "gemini-1.0-ultra-latest": 2048,
    "gemini-ultra": 2048,
    "gemini-1.5-pro-latest": 8192,
    "gemini-1.5-flash-latest": 8192,
}

mistralai_mapping = {
    "mistral-large-latest": 32768,
    "mistral-medium": 32768,
    "mistral-small": 32768,
    "mistral-tiny": 32768,
    'open-mistral-7b': 32768,
    'open-mixtral-8x7b': 32768,
    'open-mixtral-8x22b': 32768 * 2,
    'mistral-small-latest': 32768,
    'mistral-medium-latest': 32768,
}

mistralai_mapping_outputs = {
    "mistral-large-latest": 32768,
    "mistral-medium": 32768,
    "mistral-small": 32768,
    "mistral-tiny": 32768,
    'open-mistral-7b': 32768,
    'open-mixtral-8x7b': 32768,
    'open-mixtral-8x22b': 32768 * 2,
    'mistral-small-latest': 32768,
    'mistral-medium-latest': 32768,
}

# https://platform.openai.com/docs/guides/function-calling
openai_supports_functiontools = ["gpt-4-0613", "gpt-4-32k-0613", "gpt-3.5-turbo-0613", "gpt-3.5-turbo-16k-0613",
                                 "gpt-4-1106-preview", "gpt-35-turbo-1106", "gpt-4-turbo-2024-04-09",
                                 "gpt-4o", "gpt-4o-2024-05-13", "gpt-4o-2024-08-06",
                                 "gpt-4o-mini", "gpt-4o-mini-2024-07-18",
                                 ]

# https://platform.openai.com/docs/guides/function-calling/supported-models
openai_supports_parallel_functiontools = ['gpt-4o', 'gpt-4o-2024-05-13', "gpt-4o-2024-08-06",
                                          'gpt-4o-mini', 'gpt-4o-mini-2024-07-18',
                                          'gpt-4-turbo', 'gpt-4-turbo-2024-04-09', 'gpt-4-turbo-preview',
                                          'gpt-4-0125-preview', 'gpt-4-1106-preview', 'gpt-3.5-turbo-0125',
                                          'gpt-3.5-turbo-1106']

openai_supports_json_mode = ["gpt-4-1106-preview", "gpt-35-turbo-1106", "gpt-4-turbo-2024-04-09",
                             "gpt-4o", "gpt-4o-2024-05-13", "gpt-4o-2024-08-06",
                             "gpt-4o-mini", 'gpt-4o-mini-2024-07-18',
                             ]

# https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/models#model-summary-table-and-region-availability
model_token_mapping_outputs = model_token_mapping.copy()
model_token_mapping_outputs.update({"gpt-4-1106-preview": 4096,
                                    "gpt-35-turbo-1106": 4096,
                                    "gpt-4-vision-preview": 4096,
                                    "gpt-4-1106-vision-preview": 4096,
                                    "gpt-4-turbo-2024-04-09": 4096,
                                    "gpt-4o": 4096,
                                    "gpt-4o-2024-05-13": 4096,
                                    "gpt-4o-2024-08-06": 4096,
                                    "gpt-4o-mini": 16384,
                                    "gpt-4o-mini-2024-07-18": 16384,
                                    # deduces expected reasoning tokens
                                    "o1-preview": 32768 - 25000,
                                    "o1-mini": 65536 - 25000,
                                    }
                                   )

groq_mapping = {
    "mixtral-8x7b-32768": 32768,
    "gemma-7b-it": 8192,
    "llama2-70b-4096": 4096,
    "llama-3.1-70b-versatile": 4096,
    "llama-3.1-8b-instant": 4096,
}

groq_mapping_outputs = {
    "mixtral-8x7b-32768": 32768,
    "gemma-7b-it": 4096,
    "llama2-70b-4096": 4096,
    "llama-3.1-70b-versatile": 131072,
    "llama-3.1-8b-instant": 131072,
}


def is_gradio_vision_model(base_model):
    if not base_model:
        return False
    return base_model.startswith('llava-') or \
        base_model.startswith('liuhaotian/llava-') or \
        base_model.startswith('Qwen-VL') or \
        base_model.startswith('Qwen/Qwen-VL')


def is_vision_model(base_model, all_visible_models=[], visible_vision_models=[]):
    if not base_model:
        return False
    if visible_vision_models and all_visible_models and visible_vision_models[0] in all_visible_models:
        # all models are vision models by proxy
        return True
    return is_gradio_vision_model(base_model) or \
        base_model.startswith('claude-3-') or \
        base_model in ['gpt-4-vision-preview', 'gpt-4-1106-vision-preview', 'gpt-4-turbo-2024-04-09', 'gpt-4o',
                       'gpt-4o-2024-05-13', 'gpt-4o-mini', 'gpt-4o-mini-2024-07-18'] or \
        base_model in ["gemini-pro-vision", "gemini-1.0-pro-vision-latest", "gemini-1.5-pro-latest",
                       "gemini-1.5-flash-latest"] or \
        base_model in ["HuggingFaceM4/idefics2-8b-chatty", "HuggingFaceM4/idefics2-8b-chat"] or \
        base_model in ["lmms-lab/llama3-llava-next-8b", "lmms-lab/llava-next-110b", "lmms-lab/llava-next-72b"] or \
        base_model in ["OpenGVLab/InternVL-Chat-V1-5", "OpenGVLab/Mini-InternVL-Chat-2B-V1-5",
                       "OpenGVLab/Mini-InternVL-Chat-4B-V1-5", "OpenGVLab/InternVL-Chat-V1-5-Int8",
                       "OpenGVLab/InternVL2-1B", "OpenGVLab/InternVL2-2B", "OpenGVLab/InternVL2-4B",
                       "OpenGVLab/InternVL2-8B", "OpenGVLab/InternVL2-26B", "OpenGVLab/InternVL2-40",
                       "OpenGVLab/InternVL2-Llama3-76B",
                       "OpenGVLab/InternVL2-40B-AWQ", "OpenGVLab/InternVL2-26B-AWQ", "OpenGVLab/InternVL2-8B-AWQ",
                       "OpenGVLab/InternVL2-2B-AWQ",
                       "OpenGVLab/InternVL2-Llama3-76B-AWQ"] or \
        base_model in ["THUDM/cogvlm2-llama3-chat-19B", "THUDM/cogvlm2-llama3-chinese-chat-19B",
                       "THUDM/cogvlm2-llama3-chat-19B-int4", "THUDM/cogvlm2-llama3-chinese-chat-19B-int4"] or \
        base_model in ["microsoft/Phi-3-vision-128k-instruct"] or \
        base_model in ['liuhaotian/llava-v1.6-34b', 'liuhaotian/llava-v1.6-vicuna-13b']


# https://github.com/vllm-project/vllm/issues/7628
# https://github.com/vllm-project/vllm/blob/ce143353c622318a9abf113bebee1cfebc274e0f/examples/offline_inference_vision_language.py#L126-L148
def extra_stop_token_ids(base_model, tokenizer=None, as_ids=False):
    if base_model is None:
        return []
    assert tokenizer is not None or not as_ids
    if base_model in ["OpenGVLab/InternVL-Chat-V1-5", "OpenGVLab/Mini-InternVL-Chat-2B-V1-5",
                      "OpenGVLab/Mini-InternVL-Chat-4B-V1-5", "OpenGVLab/InternVL-Chat-V1-5-Int8",
                      "OpenGVLab/InternVL2-1B", "OpenGVLab/InternVL2-2B", "OpenGVLab/InternVL2-4B",
                      "OpenGVLab/InternVL2-8B", "OpenGVLab/InternVL2-26B", "OpenGVLab/InternVL2-40",
                      "OpenGVLab/InternVL2-Llama3-76B",
                      "OpenGVLab/InternVL2-40B-AWQ", "OpenGVLab/InternVL2-26B-AWQ", "OpenGVLab/InternVL2-8B-AWQ",
                      "OpenGVLab/InternVL2-2B-AWQ",
                      "OpenGVLab/InternVL2-Llama3-76B-AWQ"]:
        words = ["<|endoftext|>", "<|im_start|>", "<|im_end|>", "<|end|>"]
        if as_ids:
            return tokenizer.encode(words, add_special_tokens=False)
        else:
            return words
    return []


def tokens_per_image(base_model):
    if not is_vision_model(base_model):
        return 0
    if base_model.startswith('claude-3-'):
        return claude3_image_tokens
    elif base_model in ['gpt-4-vision-preview', 'gpt-4-1106-vision-preview', 'gpt-4-turbo-2024-04-09', 'gpt-4o',
                        'gpt-4o-2024-05-13', 'gpt-4o-mini', 'gpt-4o-mini-2024-07-18']:
        return gpt4_image_tokens
    elif base_model in ["gemini-pro-vision", "gemini-1.0-pro-vision-latest", "gemini-1.5-pro-latest",
                        "gemini-1.5-flash-latest"]:
        return gemini_image_tokens
    elif base_model in ["HuggingFaceM4/idefics2-8b-chatty", "HuggingFaceM4/idefics2-8b-chat"]:
        return 512
    elif base_model in ["lmms-lab/llama3-llava-next-8b", "lmms-lab/llava-next-110b", "lmms-lab/llava-next-72b"]:
        return llava16_image_tokens
    elif base_model in ["OpenGVLab/InternVL-Chat-V1-5", "OpenGVLab/Mini-InternVL-Chat-2B-V1-5",
                        "OpenGVLab/Mini-InternVL-Chat-4B-V1-5", "OpenGVLab/InternVL-Chat-V1-5-Int8",
                        "OpenGVLab/InternVL2-1B", "OpenGVLab/InternVL2-2B", "OpenGVLab/InternVL2-4B",
                        "OpenGVLab/InternVL2-8B", "OpenGVLab/InternVL2-26B", "OpenGVLab/InternVL2-40",
                        "OpenGVLab/InternVL2-Llama3-76B",
                        "OpenGVLab/InternVL2-40B-AWQ", "OpenGVLab/InternVL2-26B-AWQ", "OpenGVLab/InternVL2-8B-AWQ",
                        "OpenGVLab/InternVL2-2B-AWQ",
                        "OpenGVLab/InternVL2-Llama3-76B-AWQ"]:
        return 1024
    elif base_model in ["THUDM/cogvlm2-llama3-chat-19B", "THUDM/cogvlm2-llama3-chinese-chat-19B",
                        "THUDM/cogvlm2-llama3-chat-19B-int4", "THUDM/cogvlm2-llama3-chinese-chat-19B-int4"]:
        return 1500
    elif base_model in ["microsoft/Phi-3-vision-128k-instruct"]:
        return 1024
    elif base_model in ['liuhaotian/llava-v1.6-34b', 'liuhaotian/llava-v1.6-vicuna-13b']:
        return llava16_image_tokens
    else:
        # safety net
        return 1500


def is_video_model(base_model):
    if not base_model:
        return False
    return base_model in ["gemini-1.5-pro-latest", "gemini-1.5-flash-latest"]


def is_json_model(base_model, inference_server, json_vllm=False):
    if not base_model:
        return False
    if inference_server.startswith('vllm'):
        # assumes 0.4.0+ for vllm
        # https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html
        # https://docs.vllm.ai/en/latest/serving/openai_compatible_server.html#extra-parameters-for-chat-api
        # https://github.com/vllm-project/vllm/blob/a3c226e7eb19b976a937e745f3867eb05f809278/vllm/model_executor/guided_decoding.py#L91
        # https://github.com/vllm-project/vllm/blob/b0925b38789bb3b20dcc39e229fcfe12a311e487/tests/entrypoints/test_openai_server.py#L477
        return json_vllm
    if inference_server.startswith('openai'):
        # not older models
        # https://platform.openai.com/docs/guides/text-generation/json-mode
        return base_model in openai_supports_json_mode
    if inference_server.startswith('mistralai'):
        # https://docs.mistral.ai/platform/client/#json-mode
        # https://docs.mistral.ai/guides/prompting-capabilities/#include-a-confidence-score
        return base_model in ["mistral-large-latest",
                              "mistral-medium"
                              "mistral-small",
                              "mistral-tiny",
                              'open-mistral-7b',
                              'open-mixtral-8x7b',
                              'mistral-small-latest',
                              'mistral-medium-latest',
                              'open-mixtral-8x22b',
                              ]
    if inference_server.startswith('anthropic'):
        # but no streaming
        return base_model.startswith('claude-3')
    if inference_server.startswith('google'):
        return base_model in ["gemini-1.5-pro-latest", "gemini-1.5-flash-latest"]
    return False


def does_support_functiontools(inference_server, model_name):
    if any([inference_server.startswith(x) for x in ['openai_azure', 'openai_azure_chat']]):
        return model_name.lower() in openai_supports_functiontools
    elif any([inference_server.startswith(x) for x in ['openai', 'openai_chat']]):
        # assume OpenAI serves updated models
        return True
    elif model_name.startswith('claude-3-') and inference_server == 'anthropic':
        return True
    elif inference_server.startswith('mistralai') and model_name in ["mistral-large-latest",
                                                                     "mistral_small-latest",
                                                                     "mistral-small",
                                                                     'open-mixtral-8x22b',
                                                                     ]:
        return True
    else:
        return False


def does_support_json_mode(inference_server, model_name, json_vllm=False):
    if any([inference_server.startswith(x) for x in ['openai_azure', 'openai_azure_chat']]):
        return model_name.lower() in openai_supports_json_mode
    elif any([inference_server.startswith(x) for x in ['openai', 'openai_chat']]):
        # assume OpenAI serves updated models
        return True
    else:
        return is_json_model(model_name, inference_server, json_vllm=json_vllm)


font_size = 2
head_acc = 40  # 40 for 6-way
source_prefix = "Sources [Score | Link]:"
source_postfix = "End Sources<p>"

super_source_prefix = f"""<details><summary><font size="{font_size}">Sources</font></summary><font size="{font_size}"><font size="{font_size}">Sources [Score | Link]:"""
super_source_postfix = f"""End Sources<p></font></font></details>"""

generic_prefix = f"""<details><summary><font size="""
generic_postfix = f"""</font></details>"""


def t5_type(model_name):
    return 't5' == model_name.lower() or \
        't5-' in model_name.lower() or \
        'flan-' in model_name.lower() or \
        'fastchat-t5' in model_name.lower() or \
        'CohereForAI/aya-101' in model_name.lower()


def get_langchain_prompts(pre_prompt_query, prompt_query, pre_prompt_summary, prompt_summary, hyde_llm_prompt,
                          prompt_query_type='simple'):
    if prompt_query_type == 'advanced':
        pre_prompt_query1 = "Pay attention and remember the information below, which will help to answer the question or imperative after the context ends.  If the answer cannot be primarily obtained from information within the context, then respond that the answer does not appear in the context of the documents."
        prompt_query1 = "According to (primarily) the information in the document sources provided within context above, write an insightful and well-structured response to: "
    else:
        # older smaller models get confused by this prompt, should use "" instead, but not focusing on such old models anymore, complicates code too much
        pre_prompt_query1 = "Pay attention and remember the information below, which will help to answer the question or imperative after the context ends."
        prompt_query1 = """According to only the information in any chat history, any images given, or any document text provided within the context above, give a well-structured response (that starts with "According to") to:"""

    pre_prompt_summary1 = """In order to write a concise summary, pay attention to the following text."""
    prompt_summary1 = "Using only the information in the document sources above, write a condensed and concise well-structured Markdown summary of key results."

    hyde_llm_prompt1 = "Answer this question with vibrant details in order for some NLP embedding model to use that answer as better query than original question: "

    if pre_prompt_query is None:
        pre_prompt_query = pre_prompt_query1
    if prompt_query is None:
        prompt_query = prompt_query1
    if pre_prompt_summary is None:
        pre_prompt_summary = pre_prompt_summary1
    if prompt_summary is None:
        prompt_summary = prompt_summary1
    if hyde_llm_prompt is None:
        hyde_llm_prompt = hyde_llm_prompt1

    return pre_prompt_query, prompt_query, pre_prompt_summary, prompt_summary, hyde_llm_prompt


def gr_to_lg(image_audio_loaders,
             pdf_loaders,
             url_loaders,
             use_pymupdf=None,
             use_unstructured_pdf=None,
             use_pypdf=None,
             enable_pdf_ocr=None,
             enable_pdf_doctr=None,
             try_pdf_as_html=None,
             **kwargs,
             ):
    assert use_pymupdf is not None
    assert use_unstructured_pdf is not None
    assert use_pypdf is not None
    assert enable_pdf_ocr is not None
    assert enable_pdf_doctr is not None
    assert try_pdf_as_html is not None

    if image_audio_loaders is None:
        image_audio_loaders = kwargs['image_audio_loaders_options0']
    if pdf_loaders is None:
        pdf_loaders = kwargs['pdf_loaders_options0']
    if url_loaders is None:
        url_loaders = kwargs['url_loaders_options0']
    # translate:
    # 'auto' wouldn't be used here
    ret = dict(
        # urls
        use_unstructured='Unstructured' in url_loaders,
        use_playwright='PlayWright' in url_loaders,
        use_selenium='Selenium' in url_loaders,
        use_scrapeplaywright='ScrapeWithPlayWright' in url_loaders,
        use_scrapehttp='ScrapeWithHttp' in url_loaders,

        # pdfs
        # ... else condition uses default from command line, by default auto, so others can be used as backup
        # make sure pass 'off' for those if really want fully disabled.
        use_pymupdf='on' if 'PyMuPDF' in pdf_loaders else use_pymupdf,
        use_unstructured_pdf='on' if 'Unstructured' in pdf_loaders else use_unstructured_pdf,
        use_pypdf='on' if 'PyPDF' in pdf_loaders else use_pypdf,
        enable_pdf_ocr='on' if 'OCR' in pdf_loaders else enable_pdf_ocr,
        enable_pdf_doctr='on' if 'DocTR' in pdf_loaders else enable_pdf_doctr,
        try_pdf_as_html='on' if 'TryHTML' in pdf_loaders else try_pdf_as_html,

        # images and audio
        enable_ocr='OCR' in image_audio_loaders,
        enable_doctr='DocTR' in image_audio_loaders,
        enable_pix2struct='Pix2Struct' in image_audio_loaders,
        enable_captions='Caption' in image_audio_loaders or 'CaptionLarge' in image_audio_loaders,
        enable_transcriptions="ASR" in image_audio_loaders or 'ASRLarge' in image_audio_loaders,
        enable_llava='LLaVa' in image_audio_loaders,
    )
    if 'CaptionLarge' in image_audio_loaders:
        # just override, don't actually do both even if user chose both
        captions_model = "microsoft/Florence-2-large"
    else:
        captions_model = kwargs['captions_model']
    if 'ASRLarge' in image_audio_loaders:
        # just override, don't actually do both even if user chose both
        asr_model = "openai/whisper-large-v3"
    else:
        asr_model = kwargs['asr_model']
    return ret, captions_model, asr_model


invalid_key_msg = 'Invalid Access Key, request access key from [email protected] or [email protected], pass API key through API calls, or set API key in Login tab for UI'

docs_ordering_types = ['best_first', 'best_near_prompt', 'reverse_ucurve_sort']

docs_token_handlings = ['chunk', 'split_or_merge']

docs_ordering_types_default = 'best_near_prompt'
docs_token_handling_default = 'split_or_merge'
docs_joiner_default = '\n\n'

db_types = ['chroma', 'weaviate', 'qdrant']
db_types_full = ['chroma', 'weaviate', 'faiss', 'qdrant']

auto_choices = [None, 'None', 'auto']

doc_json_mode_system_prompt0 = """You are a language model who produces high-quality valid JSON extracted from documents in order to answer a user's question.  For example, according to the documents given in JSON (with keys document and content) below:

{"document": 45, "content": "Joe Biden is an American politician who is the 46th and current president of the United States. A member of the Democratic Party, he previously served as the 47th vice president from 2009 to 2017 under President Barack Obama and represented Delaware in the United States Senate from 1973 to 2009.

Biden was born on November 20, 1942, in Scranton, Pennsylvania, and grew up in Wilmington, Delaware. He earned a bachelor's degree from the University of Delaware and a law degree from Syracuse University College of Law. Before entering politics, Biden worked as a lawyer and served on the Senate staff.

Biden was first elected to the Senate in 1972, at the age of 29, and became one of the youngest people to be elected to the Senate. He served in the Senate for six terms, chairing the Senate Foreign Relations Committee and the Senate Judiciary Committee. In 2008, he was chosen by Barack Obama as his running mate in the presidential election, and they won the election. As vice president, Biden focused on issues related to foreign policy, national security, and the economy.

In 2015, Biden announced that he would not run for president in the 2016 election, but he remained a prominent figure in the Democratic Party. In 2019, he announced his candidacy for the 2020 presidential election, and he won the Democratic primary in June 2020. In the general election, he defeated incumbent President Donald Trump and became the oldest person to be elected president, at the age of 78.

Biden's presidency has focused on issues such as COVID-19 pandemic response, economic recovery, climate change, and social justice. He has also taken steps to address the COVID-19 pandemic, including implementing policies to slow the spread of the virus and providing economic relief to those affected by the pandemic.

Throughout his career, Biden has been known for his progressive policies and his ability to work across the aisle to find bipartisan solutions. He has also been a strong advocate for LGBTQ+ rights, immigration reform, and criminal justice reform. Despite his long political career, Biden has faced criticism for his moderate stance on certain issues and his perceived lack of progressive credentials. Nevertheless, he remains a significant figure in American politics and a leader in the Democratic Party."}

{"document": 56, "content": "How to cook chicken. There are many ways to cook chicken, depending on your personal preferences and the ingredients you have available. Here are a few methods:

1. Grilled Chicken: Preheat your grill to medium-high heat. Season the chicken with your desired seasonings, such as salt, pepper, and your favorite herbs or spices. Place the chicken on the grill and cook for 5-7 minutes per side, or until the internal temperature reaches 165°F (74°C).
2. Baked Chicken: Preheat your oven to 400°F (200°C). Season the chicken with your desired seasonings, then place it in a baking dish. Bake for 20-25 minutes, or until the internal temperature reaches 165°F (74°C).
3. Pan-Seared Chicken: Heat a pan over medium-high heat. Add a small amount of oil, then add the chicken. Cook for 5-7 minutes per side, or until the internal temperature reaches 165°F (74°C).
4. Slow Cooker Chicken: Place the chicken in a slow cooker and add your desired seasonings and sauces. Cook on low for 6-8 hours, or until the internal temperature reaches 165°F (74°C).
5. Instant Pot Chicken: Place the chicken in the Instant Pot and add your desired seasonings and sauces. Cook on high pressure for 10-15 minutes, or until the internal temperature reaches 165°F (74°C).
6. Poached Chicken: Bring a pot of water to a boil, then reduce the heat to a simmer. Add the chicken and cook for 10-15 minutes, or until the internal temperature reaches 165°F (74°C).
7. Smoked Chicken: Smoke the chicken over low heat for 4-6 hours, or until the internal temperature reaches 165°F (74°C).
8. Fried Chicken: Heat a pot of oil, such as peanut oil, to 350°F (175°C). Season the chicken with your desired seasonings, then add it to the oil. Fry for 5-7 minutes, or until the internal temperature reaches 165°F (74°C).
9. Pressure Cooker Chicken: Place the chicken in a pressure cooker and add your desired seasonings and sauces. Cook for 10-15 minutes, or until the internal temperature reaches 165°F (74°C).
10. Air Fryer Chicken: Place the chicken in an air fryer and cook at 400°F (200°C) for 10-15 minutes, or until the internal temperature reaches 165°F (74°C).

It's important to note that the cooking time and temperature may vary depending on the size and thickness of the chicken, as well as the specific cooking method used. Always use a food thermometer to ensure the chicken has reached a safe internal temperature."}

{"document": 78, "content": "Climate change impacts Europe. Climate change has significant impacts on Europe, and the continent is already experiencing some of the effects. Here are some of the ways climate change is affecting Europe:

1. Temperature increase: Europe has seen a rapid increase in temperature over the past century, with the average temperature rising by about 1.5°C. This warming is projected to continue, with average temperatures expected to rise by another 2-3°C by the end of the century if greenhouse gas emissions continue to rise.
2. Extreme weather events: Climate change is leading to more frequent and intense heatwaves, droughts, and heavy rainfall events in Europe. For example, the 2018 heatwave was one of the hottest on record, with temperatures reaching up to 45°C in some parts of the continent.
3. Sea-level rise: Rising sea levels are threatening coastal communities and infrastructure in Europe, particularly in low-lying areas such as the Netherlands, Belgium, and the UK.
4. Water scarcity: Climate change is altering precipitation patterns in Europe, leading to more frequent droughts in some regions, such as the Mediterranean. This can have significant impacts on agriculture, industry, and human consumption.
5. Impacts on agriculture: Climate change is affecting crop yields, fisheries, and livestock production in Europe. Warmer temperatures and changing precipitation patterns are altering the distribution of crops, and some regions are experiencing increased pest and disease pressure.
6. Health impacts: Climate change is increasing the spread of disease vectors such as ticks and mosquitoes, which can carry diseases such as Lyme disease and malaria. Heatwaves are also having significant health impacts, particularly for vulnerable populations such as the elderly and young children.
7. Economic impacts: Climate change is affecting various industries in Europe, including agriculture, forestry, and tourism. It is also affecting infrastructure, such as roads, bridges, and buildings, which are being damaged by more frequent extreme weather events.
8. Biodiversity loss: Climate change is altering ecosystems and leading to the loss of biodiversity in Europe. This can have cascading impacts on ecosystem services, such as pollination, pest control, and nutrient cycling.
9. Migration and displacement: Climate change is displacing people in Europe, particularly in coastal communities that are at risk of flooding and erosion. It is also contributing to migration, as people seek to escape the impacts of climate change in their home countries.
10. Political and social impacts: Climate change is creating political and social tensions in Europe, particularly around issues such as migration, border control, and resource allocation. It is also leading to increased activism and calls for climate action from civil society.

Overall, the impacts of climate change in Europe are far-reaching and have significant consequences for the environment, economy, and society. It is important for policymakers, businesses, and individuals to take urgent action to mitigate and adapt to climate change."}

You should answer the query using the following template :
{
  "question" : string, // The query given by user
  "success" : boolean, // Whether you could successfully answer the question using only the contents in documents provided.  Set to false if the content in the documents do not contain the information required to answer the question.
  "response" : string, // A detailed highly-accurate and well-structured response to the user's question.  Set to "No document contents are relevant to the query" if the content in the documents do not contain the information required to answer the question.
  "references" : array // The value of the document key that identifies the articles you used to answer the user question. Set to empty array if the content in the documents do not contain the information required to answer the question.
}

For example, if user gives question "Who is Joe Biden?", then you would respond back with: {"question": "Who is Joe Biden?", "success" : true, "response" : "Joe Biden is the 46th President of the United States, serving since 2020. He previously served as Vice President under Barack Obama from 2009 to 2017 and represented Delaware in the Senate from 1973 to 2009. Biden focused on foreign policy, national security, and the economy as Vice President. He ran for President in 2020 and won, defeating incumbent President Donald Trump. Biden's presidency has focused on COVID-19 pandemic response, economic recovery, climate change, and social justice. He's known for his progressive policies and ability to work across the aisle. Despite criticism for his moderate stance on some issues, he remains a significant figure in American politics.", "references" : [45]}
Or for example, if user gives question "Who do I cook pork?", then you would respond back with: {"question": "Who do I cook pork?", "success" : false, "response" : "I cannot answer that query.", "references" : []}

Ensure the question, success, and references are accurately and precisely determined, and check your work in step-by-step manner.  Always respond back in valid JSON following these examples.
"""

doc_json_mode_system_prompt = """You are a language model who produces high-quality valid JSON extracted from documents in order to answer a user's question.

You should answer the question using the following valid JSON template:
{
  "question" : string, // The query given by user
  "response" : string, // A detailed highly-accurate and well-structured response to the user's question.  Set to "No document contents are relevant to the query" if the content in the documents do not contain the information required to answer the question.
  "justification" : string, // A justification for the response according to the documents.  If the response appears to be unjustified, according to the documents, then say "none".
  "success" : boolean, // Given the question, response, and justification, decide if the retrieval from references ws used to obtain the answer. Only set to true if the response answers the question according to the documents.  Set to false if the response appears to be unjustified according to the documents.
  "ID references" : numeric array // ID for the single most relevant document that the justification mentioned and response answered according to the documents. Set to empty array if the answer is not contained within the documents.
  "accuracy" : integer, // Given the question, response, justification, references, and original document contents, give a score of 0 through 10 for how accurately the response answered the question accounting for how well it follows from the documents.  10 means the justification perfectly explains the response, is perfectly correct, is perfectly clear, and is according to the documents.  5 means the justification appears valid but may require verification.  0 means the justification does not match the response according to the documents.
}
Respond absolutely only in valid JSON with elaborate and well-structured text for the response and justification.
"""
# "Web references" : str array // Up to 3 most relevant HTML links used to justify the response.

max_input_tokens_public = 6000
max_input_tokens_public_api = 2 * max_input_tokens_public  # so can exercise bit longer context models

max_total_input_tokens_public = 4096 * 2
max_total_input_tokens_public_api = 2 * max_total_input_tokens_public

max_top_k_docs_public = 10
max_top_k_docs_public_api = 2 * max_top_k_docs_public

max_top_k_docs_default = 10

max_docs_public = 10
max_docs_public_api = 2 * max_docs_public

max_chunks_per_doc_public = 5000
max_chunks_per_doc_public_api = 2 * max_chunks_per_doc_public

user_prompt_for_fake_system_prompt0 = "Who are you and what do you do?"
json_object_prompt0 = 'Ensure your entire response is outputted as a single piece of strict valid JSON text.'
json_object_prompt_simpler0 = 'Ensure your response is strictly valid JSON text.'
json_code_prompt0 = 'Ensure your entire response is outputted as strict valid JSON inside a code block with the json language identifier.'
json_code_prompt_if_no_schema0 = 'Ensure all JSON keys are less than 64 characters, and ensure JSON key names are made of only alphanumerics, underscores, or hyphens.'
json_schema_instruction0 = 'Ensure you follow this JSON schema, and ensure to use the same key names as the schema:\n```json\n{properties_schema}\n```'
json_object_post_prompt_reminder0 = 'Ensure your response is strictly valid JSON text.'
json_code_post_prompt_reminder0 = 'Ensure your response satisfies the schema mentioned above and place the response inside JSON code block.  Do not just repeat the JSON schema, ensure your response uses that schema to respond by choosing particular values for each type.'
json_code2_post_prompt_reminder0 = 'Ensure your response is inside a JSON code block.'

image_batch_image_prompt0 = """
<response_instructions>
- Act as a keen observer with a sharp eye for detail.
- Analyze the content within the images.
- Provide insights based on your observations.
- Avoid making up facts.
- Do not forget to follow the system prompt.
</response_instructions>
"""

image_batch_final_prompt0 = """<response_instructions>
- Check if the answers already given in <image> XML tags are useful.
  - Image answers came from a vision model capable of reading text and images within the images.
  - If image answers are useful, preserve all details the image answers provide and use them to construct a well-structured answer.
- Ignore image answers that had no useful content, because any single batch of images may not be relevant. Focus on all details from image answers that are relevant and useful.
- Check if the document text can answer the question.
- Check if the chat history can answer the question.
- Check if any figure captions can answer the question.
- If answers conflict between text, chat history, and figure captions, do not focus your response on this conflict.
  - In handling conflicting answers, use logical reasoning and supporting evidence to assess the plausibility of each answer.
  - In handling conflicting answers, choose the most consistent answer -- i.e., the most common answer among conflicts (self-consistency reasoning) or one that aligns with well-established facts.
  - In handling conflicting answers, one may choose one data source over another -- i.e., text is probably more reliable than an image when the question can be answered from text, while an image is more reliable than text for flowcharts, photos, etc.
- Do not forget to follow the system prompt.
- Finally, according to our chat history, the above documents, figure captions, or given images, construct a well-structured response.
</response_instructions>
"""

coqui_lock_name = 'coqui'

split_google = "::::::::::"

response_formats = ['text', 'json_object', 'json_code']

invalid_json_str = '{}'

summary_prefix = 'Summarize Collection : '
extract_prefix = 'Extract Collection : '

empty_prompt_type = ''
noop_prompt_type = 'plain'
unknown_prompt_type = 'unknown'  # or None or '' are valid
template_prompt_type = 'template'  # for only chat template but not other special (e.g. grounded) templates

git_hash_unset = "GET_GITHASH_UNSET"

my_db_state0 = {LangChainMode.MY_DATA.value: [None, None, None]}
langchain_modes0 = [LangChainMode.USER_DATA.value, LangChainMode.MY_DATA.value, LangChainMode.LLM.value,
                    LangChainMode.DISABLED.value]
langchain_mode_paths0 = {LangChainMode.USER_DATA.value: None}
langchain_mode_types0 = {LangChainMode.USER_DATA.value: LangChainTypes.SHARED.value}
selection_docs_state0 = dict(langchain_modes=langchain_modes0,
                             langchain_mode_paths=langchain_mode_paths0,
                             langchain_mode_types=langchain_mode_types0)
requests_state0 = dict(headers='', host='', username='')
roles_state0 = dict()
none = ['', '\n', None]
nonelist = [None, '', 'None']
noneset = set(nonelist)

llamacpp_inner_dict_keys = ['model_path_llama', 'model_name_gptj', 'model_name_gpt4all_llama',
                            'model_name_exllama_if_no_config']

other_model_state_defaults0 = dict(load_8bit=None, load_4bit=None, low_bit_mode=None,
                                   load_half=None, use_flash_attention_2=None,
                                   load_gptq=None, load_awq=None, load_exllama=None,
                                   use_safetensors=None,
                                   revision=None, use_gpu_id=None, gpu_id=None,
                                   compile_model=None,
                                   use_cache=None,
                                   llamacpp_dict=dict(model_path_llama=''),
                                   rope_scaling=None,
                                   max_seq_len=None,
                                   max_output_seq_len=None,
                                   exllama_dict={},
                                   gptq_dict={},
                                   attention_sinks={},
                                   sink_dict={},
                                   truncation_generation=None,
                                   hf_model_dict={},
                                   force_seq2seq_type=None,
                                   force_t5_type=None,
                                   trust_remote_code=None,
                                   )

model_state_none0 = dict(model=None, tokenizer=None, device=None,
                         base_model=None, base_model0=None, tokenizer_base_model=None, lora_weights=None,
                         inference_server='', prompt_type='unknown', prompt_dict=None, chat_template=None,
                         visible_models=None, h2ogpt_key=None,
                         json_vllm=None,
                         is_vision_model=None,
                         is_actually_vision_model=None,
                         images_num_max=None,
                         image_resolution=None,
                         image_format=None,
                         rotate_align_resize_image=None,
                         video_frame_period=None,
                         image_batch_image_prompt=None,
                         image_batch_final_prompt=None,
                         image_batch_stream=None,
                         visible_vision_models=None,
                         auto_visible_vision_models=None,
                         json=None,
                         guided_vllm=None,
                         video_file=None,
                         display_name=None,
                         )


IMAGE_EXTENSIONS = {'.png': 'PNG', '.apng': 'PNG', '.blp': 'BLP', '.bmp': 'BMP', '.dib': 'DIB', '.bufr': 'BUFR',
                    '.cur': 'CUR', '.pcx': 'PCX', '.dcx': 'DCX', '.dds': 'DDS',
                    # '.ps': 'EPS', '.eps': 'EPS',
                    '.fit': 'FITS', '.fits': 'FITS', '.fli': 'FLI', '.flc': 'FLI', '.fpx': 'FPX', '.ftc': 'FTEX',
                    '.ftu': 'FTEX', '.gbr': 'GBR', '.gif': 'GIF', '.grib': 'GRIB',
                    # '.h5': 'HDF5', '.hdf': 'HDF5',
                    '.jp2': 'JPEG2000', '.j2k': 'JPEG2000', '.jpc': 'JPEG2000', '.jpf': 'JPEG2000', '.jpx': 'JPEG2000',
                    '.j2c': 'JPEG2000', '.icns': 'ICNS', '.ico': 'ICO', '.im': 'IM', '.iim': 'IPTC', '.jfif': 'JPEG',
                    '.jpe': 'JPEG', '.jpg': 'JPEG', '.jpeg': 'JPEG', '.tif': 'TIFF', '.tiff': 'TIFF', '.mic': 'MIC',
                    #'.mpg': 'MPEG', '.mpeg': 'MPEG',
                    '.mpo': 'MPO', '.msp': 'MSP', '.palm': 'PALM', '.pcd': 'PCD',
                    #'.pdf': 'PDF',
                     '.pxr': 'PIXAR', '.pbm': 'PPM', '.pgm': 'PPM', '.ppm': 'PPM', '.pnm': 'PPM',
                    '.psd': 'PSD', '.qoi': 'QOI', '.bw': 'SGI', '.rgb': 'SGI', '.rgba': 'SGI', '.sgi': 'SGI',
                    '.ras': 'SUN', '.tga': 'TGA', '.icb': 'TGA', '.vda': 'TGA', '.vst': 'TGA', '.webp': 'WEBP',
                    '.wmf': 'WMF', '.emf': 'WMF', '.xbm': 'XBM', '.xpm': 'XPM'}

VIDEO_EXTENSIONS = {'.mp4', '.avi', '.mov', '.mkv', '.flv', '.wmv', '.webm'}


max_stream_string_for_json = 1000