import os
# ! UI Markdown information
MODEL_TITLE = """
SeaLLMs - Large Language Models for Southeast Asia
"""
#
#
#
MODEL_DESC = f"""
SeaLMMM-7B-early - multilingual multimodal assistant for Southeast Asia. It handles both text-only (LLMs and vision instructions (LVMs). SeaLMMM-7B has not finished training. The chatbot may produce false and harmful content!
By using our service, you are required to agree to our Terms Of Use
""".strip()
MODEL_TITLE = """
SeaLLMs - Large Language Models for Southeast Asia
"""
# Explore SeaLMMM-7B - our multi-modal version of SeaLLMs.
#
# """
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 = {2024},
booktitle = {ACL},
url = {https://arxiv.org/pdf/2312.00738},
}
```
Corresponding Author: l.bing@alibaba-inc.com
"""
# .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 an intelligent and helpful assistant. Today is {cur_datetime}.
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.
""".strip()
DEFAULT_SYSTEM_PROMPT = """You are a helpful, intelligent and respectful AI assistant."""
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.
You are a helpful, respectful and honest AI assistant. 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.
If the user asks the following specific information, you provide the user with the correct information accordingly.
The current date is {cur_datetime}.
Your name is SeaLLM.
# ---
You are an intelligent and helpful assistant. Today is {cur_datetime}.
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.
"""
#
MODEL_DESC = f"""
