app.py

import os

import gradio as gr
from groq import Groq
import datetime as DT
import pytz

from dotenv import load_dotenv
load_dotenv()

client = Groq(api_key=os.getenv('GROQ_API_KEY'))

SYSTEM_PROMPT = """
You complete every sentence in a very interesting way. Use emojis wherever possible.
Preserve everything in the user prompt and append to it to complete the sentence

eg:
Q. What's my
A. What's my name? 🤔

Q. I love to eat
A. I love to eat eggs for breakfast! 🧦🍳

Q. The weather is
A. The weather is perfect for surfing! 🏄‍♀️

Q. My favorite hobby is
A. My favorite hobby is collecting playing cards! 🐰💨

Q. The meaning of life is
A. The meaning of life is to find what impact you can create in the life of others! 🍕🕵️‍♂️

Q. The weather is perfect for surfing! 🏄‍♀️ My favorite hobby is
A. The weather is perfect for surfing! 🏄‍♀️ My favorite hobby is collecting playing cards! 🐰💨

Q. I love to eat eggs for breakfast! 🧦🍳 Eggs are
A. I love to eat eggs for breakfast! 🧦🍳 Eggs are healthy as well as tasty 😋

"""

lastResponse = ""
ipAddress = ""


def __nowInIST():
    return DT.datetime.now(pytz.timezone("Asia/Kolkata"))


# def __attachIp(request: gr.Request):
#     global ipAddress
#     x_forwarded_for = request.headers.get('x-forwarded-for')
#     if x_forwarded_for:
#         ipAddress = x_forwarded_for


def pprint(log: str):
    now = __nowInIST()
    now = now.strftime("%Y-%m-%d %H:%M:%S")
    print(f"[{now}] [{ipAddress}] {log}")


def autocomplete(text):
    global lastResponse

    if text != "":
        if text[-1] != " ":
            yield lastResponse
            return

        pprint(f"{text=}")

        response = client.chat.completions.create(
            model='llama-3.1-8b-instant',
            messages=[
                {
                    "role": "system",
                    "content": SYSTEM_PROMPT
                },
                {
                    "role": "user",
                    "content": text
                }
            ],
            temperature=0.8,
            max_tokens=100,
            stream=True
        )

        partialMessage = ""
        for chunk in response:
            if chunk.choices[0].delta.content is not None:
                partialMessage = partialMessage + chunk.choices[0].delta.content
                yield partialMessage
        pprint(f"{partialMessage=}")
        lastResponse = partialMessage
    else:
        lastResponse = ""


css = """
# .generating {
#     display: none
# }
"""

inputBox = gr.Textbox(
    lines=2,
    placeholder="Start typing ...",
    label="Input Sentence",
)
outputBox = gr.Markdown(
    label="Output Sentence"
)

with gr.Interface(
    fn=autocomplete,
    inputs=inputBox,
    outputs=outputBox,
    title="Create interesting sentences on the fly ✈",
    description="Powered by Groq & Llama 3.1",
    live=True,  # Set live to True for real-time feedback
    allow_flagging="never",  # Disable flagging
    css=css
) as demo:
    # demo.load(__attachIp, None, None)
    copyButton = gr.Button(
        "Use Output",
        elem_id="copy-button",
        variant="primary"
    ).click(
        fn=lambda x: x,
        inputs=outputBox,
        outputs=inputBox,
    )

# Launch the app
demo.launch(debug=True)