app.py

"""
# Inference

import gradio as gr

app = gr.load(
    "google/gemma-2-2b-it",
    src = "models",
    inputs = [gr.Textbox(label = "Input")],
    outputs = [gr.Textbox(label = "Output")],
    title = "Google Gemma",
    description = "Inference",
    examples = [
        ["Hello, World."]
    ]
).launch()
"""
"""
# Pipeline

import gradio as gr
from transformers import pipeline

pipe = pipeline(model = "google/gemma-2-2b-it")

def fn(input):
    output = pipe(
        input,
        max_new_tokens = 2048
    )
    return output[0]["generated_text"]#[len(input):]

app = gr.Interface(
    fn = fn,
    inputs = [gr.Textbox(label = "Input")],
    outputs = [gr.Textbox(label = "Output")],
    title = "Google Gemma",
    description = "Pipeline",
    examples = [
        ["Hello, World."]
    ]
).launch()
"""

import gradio as gr
from huggingface_hub import InferenceClient
import os

# Initialize Hugging Face Inference Client
hf_token = os.getenv("HF_TOKEN")
client = InferenceClient(api_key=hf_token)

# Function to handle user inputs and fetch model responses
def chatbot(input_text, history=[]):
    #messages = [{"role": "user", "content": input_text}]
    messages = []
    for user_input, bot_response in history:
        messages.append({"role": "user", "content": user_input})
        messages.append({"role": "assistant", "content": bot_response})
    messages.append({"role": "user", "content": input_text})

    stream = client.chat.completions.create(
        model="google/gemma-2-2b-it", 
        messages=messages, 
        #temperature=0.5,
        #max_tokens=2048,
        #top_p=0.7,
        stream=True
    )

    # Concatenate streamed response
    bot_response = "".join(chunk.choices[0].delta.content for chunk in stream)

    # Update conversation history
    history.append((input_text, bot_response))
    return bot_response, history

# Gradio Interface
demo = gr.Interface(
    fn=chatbot, 
    inputs=["text", "state"], 
    outputs=["text", "state"], 
    title="Gemma Chatbot"
)

# Launch Gradio App
demo.launch()