app.py

import logging
import gradio as gr
import wandb
import transformers
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import peft
from peft import PeftModel
import torch

wandb.login()
wandb.init(project='journal-finetune', entity='benbankston2')


# Initialize logging
logging.basicConfig(level=logging.INFO)

base_model_id = "microsoft/phi-2"
model = AutoModelForCausalLM.from_pretrained(
    base_model_id,    
    device_map="auto",
    quantization_config=BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_compute_dtype=torch.bfloat16,
        bnb_4bit_quant_type="nf4",
    ),
    torch_dtype=torch.bfloat16,
    # FA2 does not work yet
    # attn_implementation="flash_attention_2",          
)


#model = pipeline("text-generation", model=model_name)
model = PeftModel.from_pretrained(model, "phi2-journal-finetune/checkpoint-175")
model.to("cuda")

tokenizer = AutoTokenizer.from_pretrained(base_model_id, add_bos_token=True, trust_remote_code=True, use_fast=False)
tokenizer.pad_token = tokenizer.eos_token

def generate_text(prompt):
    logging.info(f"Generating text for prompt: {prompt}")
    model_input = tokenizer(prompt, return_tensors="pt").to("cuda")#100
    #response = model(prompt, max_new_tokens=100, temperature=0.6, top_p=0.8, repetition_penalty=2.5, do_sample=True)
    response = tokenizer.decode(model.generate(
        **model_input, max_new_tokens=256, 
        repetition_penalty=1.11)[0], 
        temperature = 1,
        eos_token_id=tokenizer.pad_token,
        skip_special_tokens=True,
        early_stopping = True,
        )
    #best_response = response[0]['generated_text']
    logging.info(f"Generated text: {response}")
    return response

def message_and_history(input_text, history, feedback = None):
    """Manage message history and generate responses."""#100
    if history is None:
        history = []
    history2 = list(sum(history, ()))
    history2.append(input_text)
    input = ''.join(history2)
    output = generate_text(input_text)#input)
    history.append(("User", input_text))
    history.append(("Fizz Bot", output))
    return history, history

def setup_interface():
    with gr.Blocks(css='''
        body { font-family: 'Arial', sans-serif; background: #f1f1f1; }
        .container { max-width: 800px; margin: auto; padding: 20px; background-size: cover; background-repeat: no-repeat; border-radius: 8px; box-shadow: 0 4px 8px rgba(0,0,0,0.1); }
        h1 { color: #003F87; text-align: center; margin-bottom: 20px; }
        .gr-textbox { box-shadow: inset 0 2px 3px rgba(0,0,0,0.1); border-radius: 4px; border: 1px solid #7FB2E5; padding: 10px; width: auto; background-color: #fff; }
        .gradio-chatbox { background-color: #f0f0f0; }
        parameter-accordion .gr-accordion-title { font-weight: bold; font-size: 18px; } /* Custom CSS for accordion title */
        .gradio-chatbox-message-user { background-color: #4A90E2; color: white; }
        .gradio-chatbox-message-bot { background-color: #FFFFFF; color: black; }
        .gradio-chatbox-message { border-radius: 10px; padding: 10px; margin-bottom: 8px; }
        ''', theme ="Soft") as block:

        gr.Markdown("""
            <div style="background-image: url('https://my.wlu.edu/Images/communications/publications/graphic-identity/300-dpi-wordmark-blue.png'); 
            background-size: contain; 
            background-repeat: no-repeat; 
            background-position: center; 
            text-align: center; 
            height: 100px; 
            line-height: 100px; 
            font-size: 36px; 
            color: white; 
            font-family: Arial, sans-serif;">
            </div>
        """)
        gr.Markdown("<h1>Fizz Chatbot</h1>")
        gr.Markdown("<h6><i>Disclaimer: some information may be inaccurate</i></h6>")
        with gr.Accordion("Parameters", open=False, visible=True, elem_classes=["parameter-accordion"]) as parameter_row:
            temperature = gr.Slider(
                minimum = 0.0,
                maximum = 1.0,
                value = 0.7, 
                step=0.1,
                interactive = True,
                label="Temperature"
            )
            top_p = gr.Slider(
                minimum = 0.0,
                maximum = 1.0,
                value = 1.0, 
                step=0.1,
                interactive = True,
                label="Top P"
            )
            max_new_tokens = gr.Slider(
                minimum = 16,
                maximum = 1028,
                value = 128, 
                step= 32,
                interactive = True,
                label="Max tokens"
            )
        chatbot = gr.Chatbot(label="W&L AI")
        message = gr.Textbox(label="", placeholder="Ask me anything about W&L here...", elem_id="input_box")
        submit = gr.Button("Submit Query", elem_classes="specific_button")
        submit.click(
            fn=message_and_history,
            inputs=[message, gr.State()],
            outputs=[chatbot, gr.State()]
        )
        gr.Row([chatbot])
        gr.Row([message, submit])

    return block

app = setup_interface()
app.launch(debug=True)