app.py

import os
import json
import gradio as gr
import torch
from TeLVE.imagine import ImageCaptioningModel, load_model, generate_caption
from transformers import BertTokenizerFast
from huggingface_hub import hf_hub_download, list_repo_files

# Constants
MODELS_DIR = "./TeLVE/models"
TOKENIZER_PATH = "./TeLVE/tokenizer"
HF_REPO_ID = "outsu/TeLVE"
MODEL_STATE_FILE = "./TeLVE/model_state.json"
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

def load_model_state():
    """Load the model state from JSON file"""
    if os.path.exists(MODEL_STATE_FILE):
        with open(MODEL_STATE_FILE, 'r') as f:
            return json.load(f)
    return {"downloaded_models": []}

def save_model_state(state):
    """Save the model state to JSON file"""
    os.makedirs(os.path.dirname(MODEL_STATE_FILE), exist_ok=True)
    with open(MODEL_STATE_FILE, 'w') as f:
        json.dump(state, f)

def list_available_models():
    """List all .pth models in the models directory in reverse order"""
    if not os.path.exists(MODELS_DIR):
        return []
    models = [f for f in os.listdir(MODELS_DIR) if f.endswith('.pth')]
    return sorted(models, reverse=True)

def get_hf_model_list():
    """Get list of model files from HuggingFace repository in reverse order"""
    try:
        files = list_repo_files(HF_REPO_ID)
        models = [f.split('/')[-1] for f in files if f.startswith('models/') and f.endswith('.pth')]
        return sorted(models, reverse=True)
    except Exception as e:
        print(f"Error fetching models from HuggingFace: {str(e)}")
        return []

def download_missing_models():
    """Download missing models from HuggingFace with state management"""
    if not os.path.exists(MODELS_DIR):
        os.makedirs(MODELS_DIR)
    
    state = load_model_state()
    downloaded_models = set(state["downloaded_models"])
    local_models = set(list_available_models())
    hf_models = set(get_hf_model_list())
    
    # Check for models that need downloading
    models_to_download = (hf_models - local_models) - downloaded_models
    
    for model in models_to_download:
        try:
            print(f"Downloading missing model: {model}")
            hf_hub_download(
                repo_id=HF_REPO_ID,
                filename=f"models/{model}",
                local_dir=os.path.dirname(MODELS_DIR),
                local_dir_use_symlinks=False
            )
            downloaded_models.add(model)
        except Exception as e:
            print(f"Error downloading {model}: {str(e)}")
            continue
    
    # Update state with newly downloaded models
    state["downloaded_models"] = list(downloaded_models)
    save_model_state(state)

def verify_model_integrity():
    """Verify that all models in state actually exist"""
    state = load_model_state()
    local_models = set(list_available_models())
    state["downloaded_models"] = list(set(state["downloaded_models"]) & local_models)
    save_model_state(state)

def generate_description(image, model_name):
    """Generate image caption using selected model"""
    try:
        # Load model and tokenizer
        model_path = os.path.join(MODELS_DIR, model_name)
        if not os.path.exists(model_path):
            return "Error: Selected model file not found."
        
        if not os.path.exists(TOKENIZER_PATH):
            return "Error: Tokenizer not found. Please make sure you have trained a model first."

        model = load_model(model_path)
        tokenizer = BertTokenizerFast.from_pretrained(TOKENIZER_PATH)

        # Generate caption
        caption = generate_caption(model, image, tokenizer)
        return caption

    except Exception as e:
        return f"Error occurred: {str(e)}"

# Create Gradio interface
def create_interface():
    available_models = list_available_models()
    if not available_models:
        return gr.Interface(
            fn=lambda x: "No models found in ./models directory. Please train a model first.",
            inputs="image",
            outputs="text",
            title="TeLVE - Turkish efficient Language Vision Engine 🧿",
            description="Error: No models available"
        )

    interface = gr.Interface(
        fn=generate_description,
        inputs=[
            gr.Image(type="filepath", label="Upload Image"),
            gr.Dropdown(choices=available_models, label="Select Model", value=available_models[0])
        ],
        outputs=gr.Textbox(label="Generated Caption"),
        title="TeLVE - Turkish efficient Language Vision Engine 🧿",
        description="Upload an image to generate a Turkish description.",
        examples=[
            ["./images/mugla.jpg", available_models[0]],
            ["./images/example.jpg", available_models[0]]
        ] if os.path.exists("./images") else None
    )
    return interface

if __name__ == "__main__":
    print("Verifying model integrity...")
    verify_model_integrity()
    print("Checking for missing models...")
    download_missing_models()
    demo = create_interface()
    demo.launch(share=True, server_name="0.0.0.0")