multipurpose_chatbot/configs.py

import os

# ! UI Markdown information

MODEL_TITLE = """
<img src="file/seal_logo.png" style="
    max-width: 10em;
    max-height: 5%;
    height: 3em;
    width: 3em;
">
<div class="text" style="
loat: left;
padding-bottom: 2%;
">
SeaLLMs - Large Language Models for Southeast Asia
</div>
"""

# <a href='https://huggingface.co/spaces/SeaLLMs/SeaLMMM-7b'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue'></a> 
# <a href='https://huggingface.co/SeaLLMs/SeaLLM-7B-v2'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue'></a> 
# 
MODEL_DESC = f"""
<div style='display:flex; gap: 0.25rem; '>
<a href='https://github.com/damo-nlp-sg/seallms'><img src='https://img.shields.io/badge/Github-Code-success'></a>
<a href='https://huggingface.co/spaces/SeaLLMs/SeaLLM-7B'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue'></a> 
<a href='https://huggingface.co/SeaLLMs/SeaLMMM-7B-early'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue'></a> 
</div>
<span style="font-size: larger">
<a href="https://huggingface.co/SeaLLMs/SeaLMMM-7B-early" target="_blank">SeaLMMM-7B-early</a> - multilingual multimodal assistant for Southeast Asia. It handles <b>both</b> text-only (<a href="https://huggingface.co/SeaLLMs/SeaLLM-7B-v2" target="_blank">LLMs</a> and vision instructions (LVMs). <span style="color: red">SeaLMMM-7B has not finished training.</span>
</span>
<br>
<span>
<span style="color: red">The chatbot may produce false and harmful content!</span>
By using our service, you are required to agree to our <a href="https://huggingface.co/SeaLLMs/SeaLLM-Chat-13b/blob/main/LICENSE" target="_blank" style="color: red">Terms Of Use</a>
</span>
""".strip()


MODEL_TITLE = """
<img src="file/seal_logo.png" style="
    max-width: 10em;
    max-height: 5%;
    height: 3em;
    width: 3em;
">
<div class="text" style="
    float: left;
    padding-left: 2%;
    font-size: larger;
    margin-bottom: -1em;
    margin-top: -1em;
">
SeaLLMs - Large Language Models for Southeast Asia
<div style='display:flex; gap: 0.25rem; '>
<a href='https://damo-nlp-sg.github.io/SeaLLMs/'><img class="tag" src='https://img.shields.io/badge/Blog-red'></a>
<a href='https://github.com/damo-nlp-sg/seallms'><img class="tag" src='https://img.shields.io/badge/Code-success'></a>
<a href='https://huggingface.co/spaces/SeaLLMs/SeaLLM-7B-v2.5'><img class="tag" src='https://img.shields.io/badge/%F0%9F%A4%97-Spaces-blue'></a> 
<a href='https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5'><img class="tag" src='https://img.shields.io/badge/%F0%9F%A4%97-Model-blue'></a> 
<a href='https://arxiv.org/pdf/2312.00738.pdf'><img class="tag" src='https://img.shields.io/badge/Paper-red'></a>
</div>
</div>
"""

# Explore <a href="https://huggingface.co/spaces/SeaLLMs/SeaLLM-7B" target="_blank">SeaLMMM-7B</a> - our multi-modal version of SeaLLMs.
# <div style='display:flex; gap: 0.25rem; '>
# <a href='https://damo-nlp-sg.github.io/SeaLLMs/'><img class="tag" src='https://img.shields.io/badge/Blog-red'></a>
# <a href='https://github.com/damo-nlp-sg/seallms'><img class="tag" src='https://img.shields.io/badge/Code-success'></a>
# <a href='https://huggingface.co/spaces/SeaLLMs/SeaLLM-7B-v2.5'><img class="tag" src='https://img.shields.io/badge/%F0%9F%A4%97-Spaces-blue'></a> 
# <a href='https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5'><img class="tag" src='https://img.shields.io/badge/%F0%9F%A4%97-Model-blue'></a> 
# <a href='https://arxiv.org/pdf/2312.00738.pdf'><img class="tag" src='https://img.shields.io/badge/Paper-red'></a>
# </div>
MODEL_DESC = f"""
<span style="font-size: large">
<a href="https://huggingface.co/SeaLLMs/SeaLLM-7B-v2.5" target="_blank">SeaLLM-7B-v2.5</a> - a helpful assistant for Southeast Asian Languages 🇬🇧 🇻🇳 🇮🇩 🇹🇭 🇲🇾 🇰🇭 🇱🇦 🇵🇭 🇲🇲.
</span>
<br>
<span style="font-size: small">
<span style="color: red">The chatbot may produce false and harmful content!</span>
By using our service, you agree to our <a href="https://huggingface.co/SeaLLMs/SeaLLM-Chat-13b/blob/main/LICENSE" target="_blank" style="color: red">Terms Of Use</a>
</span>
""".strip()
# <span>
# <span style="color: red">NOTE: The chatbot may produce false and harmful content and does not have up-to-date knowledge.</span> 
# By using our service, you are required to agree to our <a href="https://huggingface.co/SeaLLMs/SeaLLM-Chat-13b/blob/main/LICENSE" target="_blank" style="color: red">Terms Of Use</a>, which includes 
# not to use our service to generate any harmful, inappropriate or illegal content. 
# The service collects user dialogue data for testing and improvement under 
# <a href="https://creativecommons.org/licenses/by/4.0/">(CC-BY)</a> or similar license. So do not enter any personal information!
# </span>

"""
By using our service, you are required to agree to our <a href="https://huggingface.co/SeaLLMs/SeaLLM-Chat-13b/blob/main/LICENSE" target="_blank" style="color: red">Terms Of Use</a>, which includes 
not to use our service to generate any harmful, inappropriate or illegal content. 
The service collects user dialogue data for testing and improvement under 
<a href="https://creativecommons.org/licenses/by/4.0/">(CC-BY)</a> or similar license. So do not enter any personal information!
"""

# MODEL_INFO = """
# <h4 style="display: hidden;">Model Name: {model_path}</h4>
# """
MODEL_INFO = ""

CITE_MARKDOWN = """
## Citation
If you find our project useful, hope you can star our repo and cite our paper as follows:
```
@article{damonlpsg2023seallm,
  author = {Xuan-Phi Nguyen*, Wenxuan Zhang*, Xin Li*, Mahani Aljunied*, Weiwen Xu, Hou Pong Chan, 
            Zhiqiang Hu, Chenhui Shen^, Yew Ken Chia^, Xingxuan Li, Jianyu Wang, Qingyu Tan, Liying Cheng, 
            Guanzheng Chen, Yue Deng, Sen Yang, Chaoqun Liu, Hang Zhang, Lidong Bing},
  title = {SeaLLMs - Large Language Models for Southeast Asia},
  year = 2023,
}
```
"""


# .panel-full-width.svelte-1ylopk1.svelte-1ylopk1.svelte-1ylopk1 {
CSS = """
.message-wrap.svelte-1lcyrx4>div.svelte-1lcyrx4  img {
    min-width: 200px;
    min-height: 150px;
    max-height: 600px;
    max-width; 90%;
    width: auto;
    object-fit: contain;
}
.panel-full-width.svelte-1lcyrx4.svelte-1lcyrx4.svelte-1lcyrx4 {
    padding: calc(var(--spacing-xxl) * 1);
    width: 100%
}

.panel-full-width {
    padding: calc(var(--spacing-xxl) * 1);
    width: 100%
}

img.tag {
  max-height: 1.5em;
  width: auto;
}

span.prose {
  font-size: var(--text-lg);
}
"""

USE_PANEL = bool(int(os.environ.get("USE_PANEL", "1")))
CHATBOT_HEIGHT = int(os.environ.get("CHATBOT_HEIGHT", "500"))

ALLOWED_PATHS = ["seal_logo.png"]


DEMOS = os.environ.get("DEMOS", "")

DEMOS = DEMOS.split(",") if DEMOS.strip() != "" else [
    "DocChatInterfaceDemo",
    "ChatInterfaceDemo",
    "TextCompletionDemo",
    # "RagChatInterfaceDemo",
    # "VisionChatInterfaceDemo",
    # "VisionDocChatInterfaceDemo",
]

# DEMOS=DocChatInterfaceDemo,ChatInterfaceDemo,RagChatInterfaceDemo,TextCompletionDemo



# ! server info

DELETE_FOLDER = os.environ.get("DELETE_FOLDER", "")
PORT = int(os.environ.get("PORT", "7860"))
PROXY = os.environ.get("PROXY", "").strip()

# ! backend info

BACKEND = os.environ.get("BACKEND", "debug")

# ! model information
# for RAG
RAG_EMBED_MODEL_NAME = os.environ.get("RAG_EMBED_MODEL_NAME", "sentence-transformers/all-MiniLM-L6-v2")
CHUNK_SIZE = int(os.environ.get("CHUNK_SIZE", "1024"))
CHUNK_OVERLAP = int(os.environ.get("CHUNK_SIZE", "50"))


DEFAULT_SYSTEM_PROMPT = """You are a helpful, respectful, honest and safe AI assistant."""
DEFAULT_SYSTEM_PROMPT = """You are SeaLLM, you are a helpful, respectful and honest AI assistant. Based on your internal clock, the current date time: {cur_datetime}.

Your knowledge base was last updated on August 2023. Thus, you should answers questions about events prior to and after August 2023 the way a highly informed individual in August 2023 would if they were talking to someone from the above date, and can let the human know this when relevant.

You should give concise responses to very simple questions, but provide thorough responses to more complex and open-ended questions. You should provide thorough help with writing, analysis, question answering, math, coding, and all sorts of other tasks. It uses markdown for coding."""


SYSTEM_PROMPT = os.environ.get("SYSTEM_PROMPT", DEFAULT_SYSTEM_PROMPT)

MAX_TOKENS = int(os.environ.get("MAX_TOKENS", "2048"))
TEMPERATURE = float(os.environ.get("TEMPERATURE", "0.1"))
# ! these values currently not used
FREQUENCE_PENALTY = float(os.environ.get("FREQUENCE_PENALTY", "0.0"))
PRESENCE_PENALTY = float(os.environ.get("PRESENCE_PENALTY", "0.0"))


# Transformers or vllm
MODEL_PATH = os.environ.get("MODEL_PATH", "SeaLLMs/SeaLLM-7B-v2")
MODEL_NAME = os.environ.get("MODEL_NAME", "Cool-Chatbot")
DTYPE = os.environ.get("DTYPE", "bfloat16")
DEVICE = os.environ.get("DEVICE", "cuda")

# VLLM
GPU_MEMORY_UTILIZATION = float(os.environ.get("GPU_MEMORY_UTILIZATION", "0.9"))
TENSOR_PARALLEL = int(os.environ.get("TENSOR_PARALLEL", "1"))
QUANTIZATION = str(os.environ.get("QUANTIZATION", ""))
STREAM_YIELD_MULTIPLE = int(os.environ.get("STREAM_YIELD_MULTIPLE", "1"))
# how many iterations to perform safety check on response
STREAM_CHECK_MULTIPLE = int(os.environ.get("STREAM_CHECK_MULTIPLE", "0"))

# llama.cpp
DEFAULT_CHAT_TEMPLATE = os.environ.get("DEFAULT_CHAT_TEMPLATE", "chatml")
N_CTX = int(os.environ.get("N_CTX", "4096"))
N_GPU_LAYERS = int(os.environ.get("N_GPU_LAYERS", "-1"))

# llava.llama.cpp


# Multimodal
IMAGE_TOKEN = os.environ.get("IMAGE_TOKEN", "[IMAGE]<|image|>[/IMAGE]")
IMAGE_TOKEN_INTERACTIVE = bool(int(os.environ.get("IMAGE_TOKEN_INTERACTIVE", "0")))
IMAGE_TOKEN_LENGTH = int(os.environ.get("IMAGE_TOKEN_LENGTH", "576"))
MAX_PACHES = int(os.environ.get("MAX_PACHES", "1"))


"""
# claude style
You are SeaLLM, you are a helpful, respectful and honest AI assistant. Based on your internal clock, the current date time: {cur_datetime}.

Your knowledge base was last updated on August 2023. Thus, you should answers questions about events prior to and after August 2023 the way a highly informed individual in August 2023 would if they were talking to someone from the above date, and can let the human know this when relevant.

You should give concise responses to very simple questions, but provide thorough responses to more complex and open-ended questions. You should provide thorough help with writing, analysis, question answering, math, coding, and all sorts of other tasks. It uses markdown for coding.
"""