app.py

import os
import torch
import spaces
import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
from peft.utils import get_peft_model_state_dict

# Base model and adapters
iBASE_MODEL = "mistralai/Mistral-7B-Instruct-v0.1"
ADAPTER_REPOS = {
    "witty": "ai1-test/mixtral-lora-witty",
    "charming": "ai1-test/mixtral-lora-charming",
    "sarcastic": "ai1-test/mixtral-lora-sarcastic",
    "neutral": "ai1-test/mixtral-lora-neutral",
}

HF_TOKEN = os.environ.get("HF_TOKEN")

def _get_auth_kwargs():
    return {"use_auth_token": HF_TOKEN} if HF_TOKEN else {}

# Tokeniser
tokenizer = AutoTokenizer.from_pretrained(iBASE_MODEL, **_get_auth_kwargs())
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

# LoRA state cache
_delta_cache = {}

# Persistent trait weights
current_weights = {"witty": 0.0, "charming": 0.0, "sarcastic": 0.0, "neutral": 1.0}

def apply_weights(witty, charming, sarcastic, neutral):
    vals = [0.0 if v is None else max(float(v), 0.0) for v in (witty, charming, sarcastic, neutral)]
    total = sum(vals)
    if total <= 0.0:
        normalised = [0.0, 0.0, 0.0, 1.0]
    else:
        normalised = [v / total for v in vals]
    current_weights["witty"], current_weights["charming"], current_weights["sarcastic"], current_weights["neutral"] = normalised
    return "Trait weights updated."

def load_delta(trait):
    if trait in _delta_cache:
        return _delta_cache[trait]
    repo_id = ADAPTER_REPOS[trait]
    base = AutoModelForCausalLM.from_pretrained(iBASE_MODEL, device_map="cpu", **_get_auth_kwargs())
    lora_model = PeftModel.from_pretrained(base, repo_id)
    delta = get_peft_model_state_dict(lora_model)
    _delta_cache[trait] = delta
    return delta

@spaces.GPU
def generate_response(prompt):
    weights = current_weights.copy()
    base_model = AutoModelForCausalLM.from_pretrained(
        iBASE_MODEL,
        torch_dtype=torch.float16,
        device_map="auto",
        **_get_auth_kwargs(),
    )
    merged_delta = {}
    for trait, weight in weights.items():
        delta = load_delta(trait)
        for key, tensor in delta.items():
            merged_delta[key] = merged_delta.get(key, 0) + tensor * weight
    base_model.load_state_dict(merged_delta, strict=False)
    input_ids = tokenizer(prompt, return_tensors="pt").to(base_model.device)
    with torch.no_grad():
        output_ids = base_model.generate(**input_ids, max_new_tokens=128, do_sample=False)
    response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
    if response.startswith(prompt):
        response = response[len(prompt):].lstrip("\n\r ")
    response = response.split("\n")[0].strip()
    return response

# Chat handler to maintain history
def handle_generate(prompt, history):
    if history is None:
        history = []
    reply = generate_response(prompt)
    new_history = history + [(prompt, reply)]
    log_lines = []
    for user_msg, bot_msg in new_history:
        log_lines.append(f"**You:** {user_msg}\n**Bot:** {bot_msg}")
    formatted_history = "\n\n".join(log_lines)
    return reply, new_history, formatted_history

# Build the Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("# AIVA: Customized Personality Trait Chatbot")
    gr.Markdown(
        "Enter a prompt and adjust the weights to blend witty, charming, sarcastic and neutral traits.\n"
        "Set your desired weights once, then generate multiple responses without re‑entering them."
    )

    # Trait weights
    with gr.Column():
        gr.Markdown("## Trait Weights")
        witty_input    = gr.Number(value=current_weights["witty"],    minimum=0.0, maximum=1.0, label="Witty",    precision=2)
        charming_input = gr.Number(value=current_weights["charming"], minimum=0.0, maximum=1.0, label="Charming", precision=2)
        sarcastic_input= gr.Number(value=current_weights["sarcastic"],minimum=0.0, maximum=1.0, label="Sarcastic", precision=2)
        neutral_input  = gr.Number(value=current_weights["neutral"],  minimum=0.0, maximum=1.0, label="Neutral",  precision=2)
        apply_button = gr.Button("Apply trait weights")
        status_message = gr.Markdown("")

    # Gradio UI
    with gr.Row():
        witty_slider = gr.Slider(minimum=0, maximum=1, step=0.05, value=0, label="Witty")
        charming_slider = gr.Slider(minimum=0, maximum=1, step=0.05, value=0, label="Charming")
        sarcastic_slider = gr.Slider(minimum=0, maximum=1, step=0.05, value=0, label="Sarcastic")
        neutral_slider = gr.Slider(minimum=0, maximum=1, step=0.05, value=1, label="Neutral")

    update_btn = gr.Button("Update Weights")
    output = gr.Textbox(label="Status")

    update_btn.click(
        fn=apply_weights,
        inputs=[witty_slider, charming_slider, sarcastic_slider, neutral_slider],
        outputs=output,
    )

    # Chat area
    with gr.Column():
        gr.Markdown("## Chat")
        output_box   = gr.Textbox(label="Model response", lines=6)
        chat_history_box = gr.Markdown("")
        prompt_input  = gr.Textbox(label="Your prompt", lines=2)
        generate_button = gr.Button("Generate")
        

    # Conversation state to hold the history
    conversation_state = gr.State([])

    apply_button.click(
        apply_weights,
        inputs=[witty_input, charming_input, sarcastic_input, neutral_input],
        outputs=status_message,
    )
    generate_button.click(
        handle_generate,
        inputs=[prompt_input, conversation_state],
        outputs=[output_box, conversation_state, chat_history_box],
    )

if __name__ == "__main__":
    demo.launch()