Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,129 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
from gpt4all import GPT4All
|
| 3 |
+
from huggingface_hub import hf_hub_download
|
| 4 |
+
import os
|
| 5 |
+
|
| 6 |
+
current_directory = os.getcwd()
|
| 7 |
+
model_directory = os.path.join(current_directory, "models")
|
| 8 |
+
|
| 9 |
+
title = "TaoScience"
|
| 10 |
+
|
| 11 |
+
description = """
|
| 12 |
+
<h1><center>LLM Finetuned on TaoScience<center></h1>
|
| 13 |
+
<h3><center>TaoGPT is a fine-tuned LLM on Tao Science by Dr. Rulin Xu and Dr. Zhi Gang Sha. <br> Check out- <a href='https://github.com/agencyxr/taogpt7B'>Github Repo</a> For More Information. 💬<h3><center>
|
| 14 |
+
|
| 15 |
+
"""
|
| 16 |
+
|
| 17 |
+
NOMIC = """
|
| 18 |
+
<!DOCTYPE html>
|
| 19 |
+
<html>
|
| 20 |
+
<head>
|
| 21 |
+
<title>TaoGPT - DataMap</title>
|
| 22 |
+
<style>
|
| 23 |
+
iframe {
|
| 24 |
+
width: 100%;
|
| 25 |
+
height: 600px; /* You can adjust the height as needed */
|
| 26 |
+
border: 0;
|
| 27 |
+
}
|
| 28 |
+
</style>
|
| 29 |
+
</head>
|
| 30 |
+
<body>
|
| 31 |
+
<iframe
|
| 32 |
+
src="https://atlas.nomic.ai/map/c1ce06f4-7ed0-4c02-88a4-dd3b47bdf878/f2941fb8-0f36-4a23-8cbe-40dbf76ca9e4?xs=-41.09135&xf=41.12038&ys=-22.50394&yf=23.67273"
|
| 33 |
+
></iframe>
|
| 34 |
+
</body>
|
| 35 |
+
</html>
|
| 36 |
+
"""
|
| 37 |
+
|
| 38 |
+
model_path = "models"
|
| 39 |
+
model_name = "taogpt-v1-gguf.Q5_K_M.gguf"
|
| 40 |
+
if os.path.exists(model_directory) and os.path.isdir(model_directory):
|
| 41 |
+
print("Models folder already exits")
|
| 42 |
+
else:
|
| 43 |
+
hf_hub_download(repo_id="agency888/TaoGPT-v1-GGUF-GGUF", filename=model_name, local_dir=model_path, local_dir_use_symlinks=False)
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
print("Start the model init process")
|
| 47 |
+
model = model = GPT4All(model_name, model_path, allow_download = False, device="cpu")
|
| 48 |
+
print("Finish the model init process")
|
| 49 |
+
|
| 50 |
+
model.config["promptTemplate"] = """{0}
|
| 51 |
+
"""
|
| 52 |
+
|
| 53 |
+
model.config["systemPrompt"] = "In the Context of TaoScience answer this questions: "
|
| 54 |
+
model._is_chat_session_activated = False
|
| 55 |
+
|
| 56 |
+
max_new_tokens = 2048
|
| 57 |
+
|
| 58 |
+
def generator(message, history, temperature, top_p, top_k):
|
| 59 |
+
prompt = ""
|
| 60 |
+
for user_message, assistant_message in history:
|
| 61 |
+
prompt += model.config["promptTemplate"].format(user_message)
|
| 62 |
+
|
| 63 |
+
prompt += model.config["promptTemplate"].format(message)
|
| 64 |
+
outputs = []
|
| 65 |
+
for token in model.generate(prompt=prompt, temp=temperature, top_k = top_k, top_p = top_p, max_tokens = max_new_tokens, streaming=True):
|
| 66 |
+
outputs.append(token)
|
| 67 |
+
yield "".join(outputs)
|
| 68 |
+
|
| 69 |
+
def vote(data: gr.LikeData):
|
| 70 |
+
if data.liked:
|
| 71 |
+
return
|
| 72 |
+
else:
|
| 73 |
+
return
|
| 74 |
+
|
| 75 |
+
chatbot = gr.Chatbot(bubble_full_width=False)
|
| 76 |
+
|
| 77 |
+
additional_inputs=[
|
| 78 |
+
gr.Slider(
|
| 79 |
+
label="temperature",
|
| 80 |
+
value=0.2,
|
| 81 |
+
minimum=0.0,
|
| 82 |
+
maximum=2.0,
|
| 83 |
+
step=0.05,
|
| 84 |
+
interactive=True,
|
| 85 |
+
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.",
|
| 86 |
+
),
|
| 87 |
+
gr.Slider(
|
| 88 |
+
label="top_p",
|
| 89 |
+
value=1.0,
|
| 90 |
+
minimum=0.0,
|
| 91 |
+
maximum=1.0,
|
| 92 |
+
step=0.01,
|
| 93 |
+
interactive=True,
|
| 94 |
+
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",
|
| 95 |
+
),
|
| 96 |
+
gr.Slider(
|
| 97 |
+
label="top_k",
|
| 98 |
+
value=40,
|
| 99 |
+
minimum=0,
|
| 100 |
+
maximum=1000,
|
| 101 |
+
step=1,
|
| 102 |
+
interactive=True,
|
| 103 |
+
info="limits candidate tokens to a fixed number after sorting by probability. Setting it higher than the vocabulary size deactivates this limit.",
|
| 104 |
+
)
|
| 105 |
+
]
|
| 106 |
+
|
| 107 |
+
with gr.Blocks() as demo:
|
| 108 |
+
gr.HTML("<h1><center>TaoGPTv0<center></h1>")
|
| 109 |
+
gr.HTML("<h3><center>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- <a href='https://github.com/agencyxr/taogpt7B'>Github Repo</a> For More Information. 💬<h3><center>")
|
| 110 |
+
with gr.Column():
|
| 111 |
+
with gr.Accordion(Visualise Training Data):
|
| 112 |
+
gr.HTML("<h3>Look into the dataset we used to finetune our model</h3>")
|
| 113 |
+
gr.HTML(NOMIC)
|
| 114 |
+
with gr.Column():
|
| 115 |
+
gr.ChatInterface(
|
| 116 |
+
fn = generator,
|
| 117 |
+
title=title,
|
| 118 |
+
description = description,
|
| 119 |
+
chatbot=chatbot,
|
| 120 |
+
additional_inputs=additional_inputs,
|
| 121 |
+
examples=[
|
| 122 |
+
["What is TaoScience ?"],
|
| 123 |
+
["TaoScience was written by ?"],
|
| 124 |
+
["Tell me more about TaoScience"]],)
|
| 125 |
+
RAG_Checkbox = gr.Checkbox(label="Use Retrival Augmented Generation" , value=True , interactive=False)
|
| 126 |
+
gr.Markdown("The model is prone to Hallucination and many not always be Factual")
|
| 127 |
+
|
| 128 |
+
if __name__ == "__main__":
|
| 129 |
+
demo.queue(max_size=50).launch(share=True)
|