import gradio as gr
from gpt4all import GPT4All
from huggingface_hub import hf_hub_download
import os
current_directory = os.getcwd()
model_directory = os.path.join(current_directory, "models")
title = "TaoScience"
description = """
LLM Finetuned on TaoScience
TaoGPT is a fine-tuned LLM on Tao Science by Dr. Rulin Xu and Dr. Zhi Gang Sha.
Check out- Github Repo For More Information. 💬
"""
NOMIC = """
TaoGPT - DataMap
"""
model_path = "models"
model_name = "taogpt-v1-gguf.Q5_K_M.gguf"
if os.path.exists(model_directory) and os.path.isdir(model_directory):
print("Models folder already exits")
else:
hf_hub_download(repo_id="agency888/TaoGPT-v1-GGUF-GGUF", filename=model_name, local_dir=model_path, local_dir_use_symlinks=False)
print("Start the model init process")
model = model = GPT4All(model_name, model_path, allow_download = False, device="cpu")
print("Finish the model init process")
model.config["promptTemplate"] = """{0}
"""
model.config["systemPrompt"] = "In the Context of TaoScience answer this questions: "
model._is_chat_session_activated = False
max_new_tokens = 2048
def generator(message, history, temperature, top_p, top_k):
prompt = ""
for user_message, assistant_message in history:
prompt += model.config["promptTemplate"].format(user_message)
prompt += model.config["promptTemplate"].format(message)
outputs = []
for token in model.generate(prompt=prompt, temp=temperature, top_k = top_k, top_p = top_p, max_tokens = max_new_tokens, streaming=True):
outputs.append(token)
yield "".join(outputs)
def vote(data: gr.LikeData):
if data.liked:
return
else:
return
chatbot = gr.Chatbot(bubble_full_width=False)
additional_inputs=[
gr.Slider(
label="temperature",
value=0.2,
minimum=0.0,
maximum=2.0,
step=0.05,
interactive=True,
info="Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.",
),
gr.Slider(
label="top_p",
value=1.0,
minimum=0.0,
maximum=1.0,
step=0.01,
interactive=True,
info="0.1 means only the tokens comprising the top 10% probability mass are considered. Suggest set to 1 and use temperature. 1 means 100% and will disable it",
),
gr.Slider(
label="top_k",
value=40,
minimum=0,
maximum=1000,
step=1,
interactive=True,
info="limits candidate tokens to a fixed number after sorting by probability. Setting it higher than the vocabulary size deactivates this limit.",
)
]
with gr.Blocks() as demo:
gr.HTML("TaoGPTv0
")
gr.HTML("TaoGPTv0 is a fine-tuned Mistal-7B model with a retrieval augmented generation pipeline on Tao Science by Dr. Rulin Xu and Dr. Zhi Gang Sha. Check out- Github Repo For More Information. 💬")
with gr.Column():
with gr.Accordion("Visualise Training Data"):
gr.HTML("Look into the dataset we used to finetune our model
")
gr.HTML(NOMIC)
with gr.Column():
gr.ChatInterface(
fn = generator,
title=title,
description = description,
chatbot=chatbot,
additional_inputs=additional_inputs,
examples=[
["What is TaoScience ?"],
["TaoScience was written by ?"],
["Tell me more about TaoScience"]],)
RAG_Checkbox = gr.Checkbox(label="Use Retrival Augmented Generation" , value=True , interactive=False)
gr.Markdown("The model is prone to Hallucination and many not always be Factual")
if __name__ == "__main__":
demo.queue(max_size=50).launch(share=True)