app.py

import gc, json, torch, gradio as gr
from huggingface_hub import hf_hub_download
import tiktoken

from mingpt.model import GPT 

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
REPO_ID = "to0ony/final-thesis-plotgen"

state = {"model": None, "model_name": None, "enc": tiktoken.get_encoding("gpt2")}

def load_model(model_name):
    if state["model"] is not None and state["model_name"] == model_name:
        return state["model"]

    cfg_path = hf_hub_download(repo_id=REPO_ID, filename="config.json")
    mdl_path = hf_hub_download(repo_id=REPO_ID, filename=model_name)

    with open(cfg_path, "r", encoding="utf-8") as f:
        cfg = json.load(f)

    gcfg = GPT.get_default_config()
    gcfg.model_type = None
    gcfg.vocab_size = int(cfg["vocab_size"])
    gcfg.block_size = int(cfg["block_size"])
    gcfg.n_layer    = int(cfg["n_layer"])
    gcfg.n_head     = int(cfg["n_head"])
    gcfg.n_embd     = int(cfg["n_embd"])

    model = GPT(gcfg)
    sd = torch.load(mdl_path, map_location="cpu", weights_only=False)
    model.load_state_dict(sd["model_state_dict"], strict=True)
    model.to(DEVICE)
    model.eval()

    state["model"] = model
    state["model_name"] = model_name
    return model


@torch.inference_mode()
def generate(prompt, model_choice, max_new_tokens=200, temperature=0.7, top_k=50):
    """Generiranje teksta iz prompta"""
    model = load_model(model_choice)
    enc = state["enc"]

    x = torch.tensor([enc.encode(prompt)], dtype=torch.long, device=DEVICE)

    y = model.generate(
        x,
        max_new_tokens=int(max_new_tokens),
        temperature=float(temperature),
        top_k=int(top_k) if top_k > 0 else None,
        do_sample=True
    )

    return enc.decode(y[0].tolist())


# Gradio UI
with gr.Blocks(title="🎬 PlotGen") as demo:
    gr.Markdown("## 🎬 PlotGen\nUnesi prompt i generiraj radnju filma.")

    model_choice = gr.Dropdown(
        choices=["cmu-plots-model.pt", "cmu-plots-model-enchanced.pt"],
        value="cmu-plots-model-enchanced.pt",
        label="Model"
    )
    prompt = gr.Textbox(label="Prompt", lines=5, placeholder="E.g. A young detective arrives in a coastal town...")
    max_new_tokens = gr.Slider(32, 512, value=200, step=16, label="Max new tokens")
    temperature = gr.Slider(0.1, 1.5, value=0.9, step=0.1, label="Temperature")
    top_k = gr.Slider(0, 100, value=50, step=5, label="Top-K (0 = off)")
    btn = gr.Button("Generate")
    output = gr.Textbox(label="Output", lines=15)

    btn.click(generate, [prompt, model_choice, max_new_tokens, temperature, top_k], output)

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