Ebimsv's picture
Update app.py
97bfbcb verified
raw
history blame contribute delete
No virus
6.95 kB
import gradio as gr
from ctransformers import AutoModelForCausalLM, AutoConfig, Config # import for GGUF/GGML models
import datetime
# modelfile = "models/tinyllama-1.1b-1t-openorca.Q4_K_M.gguf"
# modelfile="TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF"
modelfile="TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T"
i_temperature = 0.30
i_max_new_tokens=1100
i_repetitionpenalty = 1.2
i_contextlength=12048
logfile = 'TinyLlama.1B.txt'
print("loading model...")
stt = datetime.datetime.now()
conf = AutoConfig(Config(temperature=i_temperature,
repetition_penalty=i_repetitionpenalty,
batch_size=64,
max_new_tokens=i_max_new_tokens,
context_length=i_contextlength))
llm = AutoModelForCausalLM.from_pretrained(modelfile,
model_type="llama",
config=conf)
dt = datetime.datetime.now() - stt
print(f"Model loaded in {dt}")
def writehistory(text):
with open(logfile, 'a', encoding='utf-8') as f:
f.write(text)
f.write('\n')
f.close()
with gr.Blocks(theme='ParityError/Interstellar') as demo:
# TITLE SECTION
with gr.Row():
with gr.Column(scale=12):
gr.HTML("<center>"
+ "<h1>πŸ¦™ TinyLlama 1.1B πŸ‹ 4K context window</h2></center>")
gr.Markdown("""
**Currently Running**: [TinyLlama/TinyLlama-1.1B-Chat-v0.6](https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v0.6) &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; **Chat History Log File**: *TinyLlama.1B.txt*
- **Base Model**: TinyLlama/TinyLlama-1.1B-Chat-v0.6, Fine tuned on OpenOrca GPT4 subset for 1 epoch, Using CHATML format.
- **License**: Apache 2.0, following the TinyLlama base model.
The model output is not censored and the authors do not endorse the opinions in the generated content. Use at your own risk.
""")
gr.Image(value='imgs/TinyLlama_logo.png', width=70)
# chat and parameters settings
with gr.Row():
with gr.Column(scale=4):
chatbot = gr.Chatbot(height = 350, show_copy_button=True, avatar_images = ["imgs/user_logo.png","imgs/TinyLlama_logo.png"])
with gr.Row():
with gr.Column(scale=14):
msg = gr.Textbox(show_label=False, placeholder="Enter text", lines=2)
submitBtn = gr.Button("\nπŸ’¬ Send\n", size="lg", variant="primary", min_width=140)
with gr.Column(min_width=50, scale=1):
with gr.Tab(label="Parameter Setting"):
gr.Markdown("# Parameters")
top_p = gr.Slider(minimum=-0,
maximum=1.0,
value=0.95,
step=0.05,
interactive=True,
label="Top-p")
temperature = gr.Slider(minimum=0.1,
maximum=1.0,
value=0.30,
step=0.01,
interactive=True,
label="Temperature")
max_length_tokens = gr.Slider(minimum=0,
maximum=4096,
value=1060,
step=4,
interactive=True,
label="Max Generation Tokens")
rep_pen = gr.Slider(minimum=0,
maximum=5,
value=1.2,
step=0.05,
interactive=True,
label="Repetition Penalty")
clear = gr.Button("πŸ—‘οΈ Clear All Messages", variant='secondary')
def user(user_message, history):
writehistory(f"USER: {user_message}")
return "", history + [[user_message, None]]
def bot(history, t, p, m, r):
# SYSTEM_PROMPT = """<|im_start|>system
# You are a helpful bot. Your answers are clear and concise.
# <|im_end|>
# """
# prompt = f"<|im_start|>system<|im_end|><|im_start|>user\n{history[-1][0]}<|im_end|>\n<|im_start|>assistant\n"
SYSTEM_PROMPT = """<|im_start|>system
You are a customer support chatbot for an online platform.
Your purpose is to assist users with their inquiries and provide accurate information.
You have been trained with a knowledge base that includes rules and limitations regarding chargebacks.
The knowledge base consists of the following information:
1. Chargebacks beyond 90 days are not possible.
2. Chargebacks above $1000 are not allowed.
3. Chargebacks for transactions with a valid 3D secure are not allowed.
Use the provided conversation example as a starting point for training.
Your goal is to respond to user queries in a helpful and informative manner, ensuring that you adhere to the platform's chargeback policies.
<|im_end|>
"""
prompt = f"<|im_start|>system<|im_end|><|im_start|>user\n{history[-1][0]}<|im_end|>\n<|im_start|>assistant\n"
print(f"history lenght: {len(history)}")
if len(history) == 1:
print("this is the first round")
else:
print("here we should pass more conversations")
history[-1][1] = ""
for character in llm(prompt,
temperature = t,
top_p = p,
repetition_penalty = r,
max_new_tokens=m,
stop = ['<|im_end|>'],
stream = True):
history[-1][1] += character
yield history
writehistory(f"temperature: {t}, top_p: {p}, maxNewTokens: {m}, repetitionPenalty: {r}\n---\nBOT: {history}\n\n")
# Log in the terminal the messages
print(f"USER: {history[-1][0]}\n---\ntemperature: {t}, top_p: {p}, maxNewTokens: {m}, repetitionPenalty: {r}\n---\nBOT: {history[-1][1]}\n\n")
# Clicking the submitBtn will call the generation with Parameters in the slides
submitBtn.click(user, [msg, chatbot], [msg, chatbot], queue=False).then(bot, [chatbot,temperature,top_p,max_length_tokens,rep_pen], chatbot)
clear.click(lambda: None, None, chatbot, queue=False)
demo.queue() # required to yield the streams from the text generation
demo.launch(inbrowser=True, share=True)