app.py

import gradio as gr
import tempfile
import imageio
import torch
import time
import os

from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
from diffusers import DiffusionPipeline


# -------------------------------------------------
# Device Setup
# -------------------------------------------------

device = "cuda" if torch.cuda.is_available() else "cpu"

text_model_cache = {}
chat_memory = {}


# -------------------------------------------------
# Available Text Models
# -------------------------------------------------

AVAILABLE_MODELS = {
    "Codette LoRA (Llama-3.1)": "codette_lora",
    "Mistral-7B Instruct": "mistralai/Mistral-7B-Instruct-v0.2",
    "Phi-3 Mini": "microsoft/phi-3-mini-4k-instruct",
    "GPT-2 (lightweight)": "gpt2"
}


# -------------------------------------------------
# Load Codette LoRA Adapter
# -------------------------------------------------

def load_codette_lora():

    base_model = "meta-llama/Meta-Llama-3.1-8B"

    tokenizer = AutoTokenizer.from_pretrained(base_model)

    model = AutoModelForCausalLM.from_pretrained(
        base_model,
        torch_dtype=torch.float16 if device == "cuda" else torch.float32,
        device_map="auto"
    )

    model = PeftModel.from_pretrained(
        model,
        "Raiff1982/codette-lora-adapters"
    )

    pipe = pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        device_map="auto"
    )

    return pipe


# -------------------------------------------------
# Image Generator (SDXL Turbo)
# -------------------------------------------------

try:

    image_generator = DiffusionPipeline.from_pretrained(
        "stabilityai/sdxl-turbo",
        torch_dtype=torch.float16 if device == "cuda" else torch.float32,
        variant="fp16" if device == "cuda" else None
    )

    image_generator.to(device)

    image_enabled = True

except Exception as e:

    print(f"[Image Model Load Error]: {e}")
    image_generator = None
    image_enabled = False


# -------------------------------------------------
# Video Generator (Zeroscope)
# -------------------------------------------------

try:

    video_pipeline = DiffusionPipeline.from_pretrained(
        "cerspense/zeroscope_v2_576w",
        torch_dtype=torch.float16 if device == "cuda" else torch.float32
    )

    video_pipeline.to(device)

    video_enabled = True

except Exception as e:

    print(f"[Video Model Load Error]: {e}")
    video_pipeline = None
    video_enabled = False


# -------------------------------------------------
# Load Text Models
# -------------------------------------------------

def get_text_model(model_name):

    if model_name not in text_model_cache:

        if AVAILABLE_MODELS[model_name] == "codette_lora":

            text_model_cache[model_name] = load_codette_lora()

        else:

            text_model_cache[model_name] = pipeline(
                "text-generation",
                model=AVAILABLE_MODELS[model_name],
                device=0 if device == "cuda" else -1
            )

    return text_model_cache[model_name]


# -------------------------------------------------
# Codette Terminal Logic
# -------------------------------------------------

def codette_terminal(prompt, model_name, generate_image, generate_video,
                     session_id, batch_size, video_steps, fps):

    if session_id not in chat_memory:
        chat_memory[session_id] = []

    if prompt.lower() in ["exit", "quit"]:
        chat_memory[session_id] = []
        yield "🧠 Codette session reset.", None, None
        return

    try:

        model = get_text_model(model_name)

        result = model(
            prompt,
            max_new_tokens=200,
            temperature=0.7,
            top_p=0.9,
            do_sample=True
        )

        output = result[0]["generated_text"]

    except Exception as e:

        yield f"[Text generation error]: {e}", None, None
        return


    # -------------------------------------------------
    # Stream Text
    # -------------------------------------------------

    response_so_far = ""

    for char in output:

        response_so_far += char

        temp_log = chat_memory[session_id][:]

        temp_log.append(f"🖋️ You > {prompt}")
        temp_log.append(f"🧠 Codette > {response_so_far}")

        yield "\n".join(temp_log[-10:]), None, None

        time.sleep(0.01)


    chat_memory[session_id].append(f"🖋️ You > {prompt}")
    chat_memory[session_id].append(f"🧠 Codette > {output}")


    imgs = None
    vid = None


    # -------------------------------------------------
    # Image Generation
    # -------------------------------------------------

    if generate_image and image_enabled:

        try:

            result = image_generator(
                prompt,
                num_images_per_prompt=batch_size,
                num_inference_steps=2
            )

            imgs = result.images

        except Exception as e:

            print(f"[Image error]: {e}")


    # -------------------------------------------------
    # Video Generation
    # -------------------------------------------------

    if generate_video and video_enabled:

        try:

            result = video_pipeline(
                prompt,
                num_inference_steps=video_steps
            )

            frames = result.frames

            temp_video_path = tempfile.NamedTemporaryFile(
                suffix=".mp4",
                delete=False
            ).name

            imageio.mimsave(temp_video_path, frames, fps=fps)

            vid = temp_video_path

        except Exception as e:

            print(f"[Video error]: {e}")


    yield "\n".join(chat_memory[session_id][-10:]), imgs, vid


# -------------------------------------------------
# Gradio Interface
# -------------------------------------------------

with gr.Blocks(title="🧬 Codette Terminal") as demo:

    gr.Markdown("## 🧬 Codette Terminal")
    gr.Markdown("Chat with Codette, generate images, and create short videos.")

    session_id = gr.Textbox(value="default_session", visible=False)

    with gr.Row():

        model_dropdown = gr.Dropdown(
            choices=list(AVAILABLE_MODELS.keys()),
            value="Codette LoRA (Llama-3.1)",
            label="Language Model"
        )

    with gr.Row():

        generate_image_toggle = gr.Checkbox(
            label="Generate Image(s)",
            value=False,
            interactive=image_enabled
        )

        generate_video_toggle = gr.Checkbox(
            label="Generate Video",
            value=False,
            interactive=video_enabled
        )

    with gr.Row():

        batch_size_slider = gr.Slider(
            label="Number of Images",
            minimum=1,
            maximum=4,
            step=1,
            value=1
        )

        video_steps_slider = gr.Slider(
            label="Video Inference Steps",
            minimum=10,
            maximum=50,
            step=10,
            value=20
        )

        fps_slider = gr.Slider(
            label="Video FPS",
            minimum=4,
            maximum=24,
            step=2,
            value=8
        )

    user_input = gr.Textbox(
        label="Your Prompt",
        placeholder="A robot dreaming on Mars...",
        lines=1
    )

    output_text = gr.Textbox(
        label="Codette Output",
        lines=15,
        interactive=False
    )

    with gr.Row():

        output_image = gr.Gallery(
            label="Generated Images",
            columns=2
        )

        output_video = gr.Video(
            label="Generated Video"
        )


    user_input.submit(

        codette_terminal,

        inputs=[
            user_input,
            model_dropdown,
            generate_image_toggle,
            generate_video_toggle,
            session_id,
            batch_size_slider,
            video_steps_slider,
            fps_slider
        ],

        outputs=[
            output_text,
            output_image,
            output_video
        ]

    )


# -------------------------------------------------
# Launch
# -------------------------------------------------

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